Designing a Better Social Media Browser

							Designing a Better Social Media Browser

Final Year Undergraduate Project Jonathan Melhuish

Supervisor: Russell Beale

2008

Table of Contents 1. Introduction 2. Background Research
● ● ● ● ●

Social Media Information Overload Information Filtering Behaviour Automatic Information Filtering Social Psychology

3. User Research
● ●

● ● ● ●

Interviews Questionnaire ○ Demographics ○ Number of Friends ○ News Feed ○ Key Findings Survey of Published Items User perception of News Feed completeness Utilisation of News Feed filtering features Similarity between Friends

4. Design
● ● ● ● ●

● ● ● ● ● ● ● ● ● ●

Objectives Design Approach Analysis of social media monitoring Personas Analysis of existing interfaces ○ Analysis of social-networking interfaces ■ Facebook ○ Analysis of RSS feed-reader interfaces ■ Flock ○ Conclusions Evaluation of existing interfaces Persona-based evaluation of existing interfaces Interface concepts Design 1 Design 2 Design 3 Design 4 Design 5 Automatic filtering of social media Automatic topography generation for representing social context

5. Prototype Implementation
● ●

Specification Technical Details

● ● ● ● ● ● ●

System architecture Models – database tables Models – methods Models – validation View Controllers Testing Evaluation of Prototype

6.
● ● ●

Walkthrough of typical monitoring task Persona-based evaluation User evaluation Conclusion

7.

Please note that for brevity, supporting data is included on the CD rather than as Appendices. See Appendix 1 for details of the contents of the CD.

Abstract
Recent years have seen a rapid rise in the use of online social networking services, which assist users in discovering, viewing and publishing digital media for the purposes of communication amongst peers. This research explores the nature and usage of social networking services, then focusses on the task of selecting and viewing the digital media produced by peers. Existing User Interface designs are evaluated and alternative designs proposed, some of which are based on Machine Learning techniques. These designs are evaluated and compared, and a digital prototype is constructed of one of the designs. This is evaluated using both a persona-based approach and feedback gathered from users of social networking services. Keywords Social Media, Social Networking Service, Facebook, HCI, User Experience Design, Information Overload, Machine Learning, Social Psychology Acknowledgements Dr. Russell Beale Dr. Julie Christian Dr. Simon Hammond Dr. Allan White Paul Goscicki Simon Menashy Benjamin Harpin I would also like to thank all those who voluntarily participated in my interviews and questionnaires.

Introduction
Online social networking services have dramatically risen in popularity in recent years, particularly amongst younger people. These services facilitate multi-modal communication amongst users by making it easy to discover content created by friends and also to create, upload and publish content. This content may be textual, visual, or multimedia or may be another form of interaction between users such as playing a game against one another. It seems that many use social networking services to interact with a broad range of people, from colleagues and classmates whom they see every day to long-lost friends who they have not seen for years, and from the closest of companions to the most casual of acquaintances. Although these relationships may differ, they all seem to benefit from the rich communication tools offered by a social network. Brief exploration of the user profiles on a social networking service reveals that these relationships are not only varied but vast in number. Many users seem to have several hundred “friends” in their online social network. Talking to users quickly reveals that it is quite unusual to ever delete a friend from one's online social network, so this number will presumably grow ad infinitum. Social psychology tells us that the user's number of meaningful relationships stays relatively constant and small (around 20). This implies that a large proportion of the “friends” on the average social networking service's profile are more accurately acquaintances with which the user has no significant social bond, meaning that he may make little effort to maintain the relationship. This research seeks to investigate these hypotheses, exploring the nature of online relationships via social networking services and investigating how people make use of these online tools to support these relationships. In particular, we will focus on the “listening” part of this online discussion, finding out how users pick out news of interest amongst this hubbub of activity. We will then evaluate how User Interface design can assist in this task.

photo by Jonathan Melhuish

Background Research
Social media There is no universally-agreed definition of social media, and succinct definitions are few and far between. One of the most frequently cited definitions is Wikipedia (2008), which defines social media as “the democratization of information, transforming people from content readers into content publishers. It is the shift from a broadcast mechanism to a many-to-many model, rooted in conversations between authors, people, and peers.” For the purposes of this research, I will loosely define social media as online media that is created as part of an open discussion. This contrasts to broadcast media, where information flow is largely one-way, and with direct communication, which happens between individuals or within a closed group. Social media lies in between: items may be published publicly but are generally viewed by a certain social group who then join in a discussion of the original post. Social media can take many forms, including: ● weblogs (“blogs”) - online diaries that the author updates regularly. Blogs can be on any topic (indeed, most bloggers write about many topics) but the most common topic is the author's life and personal experiences (Lenhart & Fox, 2006). Blogs are typically created using a specialised web-based tool that allows easy publishing. ● micro-blogs – micro-blogs are also predominantly text-based, but are geared towards mobile production and consumption. This interface allows for more spontaneous messages, usually concerning the author's current situation, activity or thoughts. Many people use their name or subtitle facility on their instant messenger to serve a similar purpose. ● video blogs – it is now relatively cheap and easy to publish video content on the web. Some authors use this technology to simply to deliver similar content to a written blog, but many use it to publish documentary-style or instructional video, or audio-visual art. ● photo sharing – digital cameras make it easy to share photographs over the Internet. Many people upload photographs of social occasions, or photographic art. Social networking services typically offer tools for publishing some or all of these forms of media. They have the advantage of having a digital profile of the user's social network which can be used to display media published by a user's friends, thus acting as a form of social news aggregator, or social filter. It can also highlight media that is somehow connected to the user, such as photographs of them at a party of a video of a talk they gave. In order to get a deeper understanding of what is understood by social media, I conducted a brainstorm with Dr. Simon Hammond, who is creating a new social networking platform1 and has a comprehensive understanding of social media and online social networking. We explored the nature of “friends” in the context of social media, and made the following observations: ● the context of the relationship has a large effect on social behaviour within relationships, e.g. a person may reveal details to close friends that it would be socially inappropriate to discuss with his manager ● the location affects the nature of the relationship, mainly as it has a large effect on frequency of face-to-face contact. Friendships may also start largely due to being in a shared location or situation. Those in similar situations also have shared information requirements; this verbal exchange of information is often the start of a friendship. .
1 Bodder, a social networking service designed for mobile devices. http://www.bodder.com

●

● ●

the social proximity of the two parties is continuously variable and changes over time – it is certainly not binary and unchanging, as assumed by most online social networking platforms. it was proposed that the expected likelihood of seeing the author again in the near future would be a significant determinant of the level of interest in social information we considered that family were quite different to any group of friends, even close friends.

Our discussion of social media itself sought primarily to contrast it with other more direct forms of communication. Essentially, we consider that most social media is “broadcast” media – it is published without any closely-defined audience but with a loose expectation that it may be of interest to the friends of the author. Clearly some authors of blogs seek to reach a wider audience, but as the social proximity of author and viewer decreases, it becomes questionable whether the media is truly “social”. However, the implicit purpose of social media is to stimulate discussion, which may happen in the same medium (e.g. one blog post in response to another), switch to a more direct form of communication (e.g. telephone) or start a face-to-face discussion. The key advantage of social media is in its “broadcast” nature, that it may be easily consumed by a large number of viewers without additional effort on the part of the author. Its key disadvantage is that it is impersonal, which may mean that the communication may be far less effective and emotive than a one-to-one discussion, and that, because no viewers feel that it is directed specifically at them, they feel less compelled to respond. Information overload The term “information overload” was coined by Toffler (1970), and refers to the state of having too much information to make a decision or remain informed about a topic. Information overload can lead to increased stress levels and decreased performance. As it becomes ever easier to create and publish content, and also to access it, many people seem to not only to consume more information but also increase the variety of sources. People often find it difficult to process information from all of these various sources, but don't want to simply ignore it in case they are “left out of the loop”.

Information filtering behaviour There is very little published research specifically concerning information filtering behaviour within the context of social media, but more general frameworks have been proposed that may be applied. Sonnenwald (1998) proposes a framework that is built upon previous research and is in the form of three propositions, which I will briefly interpret in relation to social media: “Proposition 1: Human information behavior is woven around, i.e., is shaped by and shapes, individuals, social networks, situations and contexts. ” An individual's social network plays a role in identifying information needs and exploring them; in the case of social media, perhaps it is the social network itself that creates the information need, as staying informed about the lives of others is necessary for the maintenance of relationships. “Proposition 2: Individuals or systems within a particular situation and context, may perceive, reflect and/or evaluate change in others, self, and/or their environment. Information behavior is constructed amidst a flow of such reflections and/or evaluations, in particular, amidst reflections and/or evaluations concerning a lack of knowledge.” Individuals and their life stories are in a constant state of change as they perform actions, and hence if one individual has not been recently in touch with another, they may perceive that the other is likely to have changed, creating a desire for information about that person. Equally, they may perceive a considerable change in themselves and take actions to report it to others. Knowledge about the present activities of another individual may influence the perception of change, therefore causing others to be more interested in information about that individual. “Proposition 3: Within a context and situation is an “information horizon” in which we can act.” For a particular situation and context, both social networks and individuals determine the information horizon - that is, the resources that the individual may use. For example, online social networks often allow individuals to publish personal information only to those whom they have identified as a friend, so whether or not the information-seeker has friend status will determine the visibility of this information. This behaviour strikes a balance between open information sharing and maintaining users' privacy. After systematically applying Sonnewald's framework to online communities, Lin & McDonald. (2006) conclude that accessibility ought to also be interpreted in terms of normative access - the socially and individually shaped norms that approve or disprove one’s use of a resource. Even if the individual has access to information about others, it may be inappropriate to make use of it but it would be socially inappropriate for them to exploit this access. Information horizons may also be limited by the use of multiple online identities. Authors may publish under several pseudonyms, which may only be known by certain people. This effectively limits access to social information even though it is in the public domain. “Proposition 4: Human information behavior may, ideally, be viewed as collaboration among an individual and information resources.” The goal of this collaboration is to share meaning and to provide the recipient with the required

knowledge, using the resources available within that individual's information horizon. In the context of social media, this might be knowledge about aspects of another individual's life. The collaboration also ideally includes reflexive provisioning of information, such as writing a blog post concerning a topic the author knows his readers will be interested in. “Proposition 5: Information horizons may be conceptualized as densely-populated solution spaces.” As the various information resources that form an information horizon have some knowledge of each other, they can be seen as a densely-populated solution space in which the information retrieval challenge for the individual is to make solutions visible. For example, this may involve expanding an individual's social network to include those who can provide the desired information. In the case of social media, perhaps an individual might try to forge new relationships with the friends of somebody he is romantically interested in, so that he can collect information about that individual that he would not otherwise have access to. Automatic information filtering An information filter is a system that aims to increase the “signal-to-noise” ratio and help to avoid information overload. It does this by comparing the user's profile with either the item itself (content-based approaches) or with the user's social environment (collaborative filtering). The user's profile might be built from explicit feedback, typically clicking an on-screen feedback button, or by observing characteristics of their behaviour, such as how long they spend looking at the item. The former approach may lead to a more certain measure of the user's opinion, but the latter approach generates more data. Research has shown that there is a high correlation between the two sources and that using observed data may in fact lead to more accurate predictions (Morita & Shinoda, 1994). For text-based media, content-based approaches generally maintain a list of words that occur most frequently in items that the user is interested in. Similar approaches can be devised for some other data types, but it is hard to devise efficient approaches for video and audio. Sometimes textual “tags” are available that can be used, but because they are generally only a small number, it may be difficult to identify patterns that indicate the user's interest. One solution to this problem is to use a social filtering approach. By gathering feedback from a large number of users about an item, it is possible to gauge how interesting the community finds that item. This is the basis of sites like Digg2, which shows recent items that the Digg community rates highly on its front page. When feedback is available from a large number of users about a large number of items, not only can the items be rated, but the users themselves can be meaningfully grouped together, according to the similarity of their ratings to the ratings of other users for the same items. This allows personalised recommendations to be made, and is the basis of many services such as Amazon3 product recommendations and the Last.fm4 music recommendation service. A variation to this approach is to build a representation of the user's social network and instead use this to generate recommendations. Groh & Ehmig (2007) have shown this approach to be more accurate than collaborative filtering in taste-based domains.
2 Digg, a community-driven links site. http://www.digg.com 3 Amazon is a major online online retailer. http://www.amazon.com 4 Last.fm is a music recommendation service based on collaborative filtering. http://www.last.fm

By computing an estimate of how interesting a user will find an item, the system can guide the user towards items that are more likely to be of interest. Conversely, the user can choose not to view every item, safe in the knowledge that the unseen items are unlikely to be of interest. In this way, an information filter (or recommender system) can direct the user's attention toward relevant items, help them to use their time effectively, and help to reduce the stress associated with information overload. Social psychology Social media is often created to support existing relationships, so it is appropriate to give a very brief overview of the psychology of relationships. The reason that individuals form relationships is termed “interpersonal attraction” and is governed by several key factors, which may often be overlapping: ● personal characteristics – there are physical features and personality traits that individuals find attractive in others. Each individual is also more likely to be attracted to other individuals who share similar characteristics to themselves. These characteristics are generally quite hard to accurately capture empirically. ● demographics are similar to personal characteristics but are more objective, such as age, education, profession, etc. They are often an indication of an individual's attitudes and values and so affect interpersonal attraction. ● familiarity – how much exposure people have had to each other has a strong effect on personal attraction (Zajonc, 1968). Online communication can form strong relationships, but the process tends to take longer than with regular face-to-face contact. (Parks & Roberts 1998). The social context in which the relationship is formed also affects how it develops. The attributes that affect how likely the relationship is to form and develop are different in different contexts (Jackson 1977). The concept of social structure is likely to be important in modelling relationships. A common structure is that of the social network, which is a graph that represents people as nodes and relationships as arcs. Mathematical graph theory can be applied to represent and analyse factors such as the strength of social ties, roles, information flow, etc. This approach can be used to analyse a whole social network (within some boundaries), or with an “egocentric” approach that seeks to view the social network from the perspective of a single individual's relationships. Analysing from a whole network perspective allows the exploration of the social network structure, composition and functioning (Garton, 2008). Social psychology research identifies two common social groups – the “sympathy group” and the “support clique”. A sympathy group corresponds to the number of people contacted by an individual at least once a month, and is generally 12-20 people (Dunbar & Spoors, 1995). Men and women do not differ in their total network size, but women have more females and more kin in their networks than men do. A “support clique” is typically 3-5 individuals from whom the respondent would seek personal advice or help in times of severe financial or emotional distress (Zhou et al., 2005). Each sex exhibits a strong preference for members of their own sex (Dunbar & Spoors, 1995).

User research
Interviews In order to explore the context and manner in which people consume social media, I conducted a series of interviews with users of social networking services of different ages, nationalities and levels of computer expertise. The interviewees were volunteers from amongst the author's family and friends. Interview 1 was conducted face-to-face, all other interviews were conducted by telephone. In interview 3, I was able to observe the interviewee's computer screen. Please see the CD for a copy of the interview notes. # 1 2 3 4 5 Name Mike Kim Hania Marianne Anka Age Early 20s Early 50s Mid 20s Mid 20s Mid 20s Typical computer usage Programming, research Email, online shopping, word processing Web surfing, email, digital photography Web surfing, email, digital photography Web surfing, email

All the interviewees saw social media as primarily useful for supporting existing relationships, predominantly with their closer friends. In general, they believe that it rarely starts new relationships, nor does much to revive old ones. However, social media, particularly in the context of social networking services, seems to strengthen relationships by exposing aspects of an author's life that the viewer may not be aware of, and are perhaps outside the context in which the relationship began. One interviewee reported that online social networks provide an easy way of staying touch with friends. This may be because interactions are usually quicker and more casual than email or telephone communications. Also there is much greater opportunity for passive browsing of information about other individuals' lives that they have published online, which often initiates and supports these casual social interactions. Two users, both of whom use social networking services regularly, reported that they only view a small fraction of the social media that their friends produce. One user suggested that the amount of social media presented to him was increasing mostly because he continually adds more friends to his online social network. While all users generally consume a wide variety of media types, they report a strong preference towards items that do not take long to consume, such as photographs and microblogs. Perhaps this is because they feel that the time investment in longer items might not be rewarded, or that they simply lack the commitment to consume longer items. One interviewee discussed how she had very little interest in content that was created as the result of a friend's action, e.g. adding a new friend to their online social network or using an application. Interviewees generally reported that created digital media for “their friends”. None of the interviewees specifically identified any target individual or social group. The interviewees all perceived a strong link between their social proximity to the author or subject of the story and their level of interest in the media. Most interviewees state that use social networking services to stay informed about the lives of people that they see regularly as well as, in combination with other forms of telecommunication, to maintain relationships with those with whom they were close and saw regularly, but are now separated by distance.

A key indication of social proximity appears to be the likelihood or desire to meet this friend in future. With the exception of family, the frequency, duration and “quality” of offline contact appears to be important in determining how close they felt to a particular person. This is supported by the theory proposed by Zajonc (1968) that states that any exposure to any mildly negative, neutral or positive stimulus leads to an increasingly positive evaluation of that stimulus. In this case, this means that unless an individual takes a particular dislike to someone, they will tend to react more positively to them each time they meet. One interviewee suggested that if she was suddenly able to only see the output of her ten closest friends, she would feel no real loss. However, many respondents reported that they regularly browsed the online profiles of people they do not know, or know very little. Several users termed this “spying”, and suggested that it is the product of the natural inclination to have an interest in the lives of others. Whilst I suggest this behaviour may be important in converting acquaintances into friends, the interviewees did not report it as such. It is not clear whether the subjects could classify their friends into a small number of “levels of friendship” or whether they perceive a smooth continuum of social proximity. One interviewee considered that she could split her friends into groups of differing social proximity easily. However, I strongly suspect that this is because her mother tongue (Polish) has different words for different levels of friendship (przyjaciele for the closest of friends, koledzy/koleżanki for regular friends and znajomi for casual acquaintances), so she is frequently required to classify her friends in this way. This ability may therefore not extend to speakers of other languages. What is clear, however, is that people consider some friends to be closer than others, and that this has a strong influence on how much attention they pay to that person's news. I argue therefore that by making it easy to discover and add new “friends”, but failing to distinguish between close friends and casual acquaintances, social networking services will present their users with an ever-increasing amount of media. As the amount presented to them increases, an ever-decreasing proportion will be considered interesting. This will force them to: spend longer filtering the information; remove friends from their online social network (an action which appears to be viewed as a social faux pas); pay more attention to sites on which they have only close friends; or perhaps, as the level of interest decreases below some threshold, ignore it completely. Questionnaire After completing the interviews, I decided that it would be worthwhile to conduct an online survey to confirm my interpretation of the interview results. Draft versions of the questionnaire were tested on a small number of respondents first (whose responses are not included here), with some questions being re-worded as a result of their feedback. See the CD for a copy of the final survey questions. The survey focussed on: ● how often and for how long people use social networking services ● for which communication tasks people consider social networking services most important ● how people use aggregated news feeds, or how they monitor their friends' activities if they do not use the aggregated feed, and why ● which items of news they found most interesting ● their friends that are part of their online social network The survey also attempted to examine the relationship between the latter two items, by

measuring: ● how many friends they had in several categories, and in total ● how interested they are in hearing news relating to people in each of these categories The survey was sent to the author's friends on Facebook, a popular social networking platform. The results are discussed below. Where a correlation between two variables is given, this indicates that a Pearson Correlation was found with a significance level of at least 0.05. In the case of comparisons between the ratings given to different categories, the Wilcoxon Signed Rank Test was used. As multiple hypotheses (pair-wise comparisons) were being tested, the probability of finding a hypothesis to be true by chance is increased. A higher significance threshold of 0.01 was used to compensate for this. The statistics were analysed using SPSS. Demographics 148 of the author's Facebook friends were asked to participate in the study, of whom precisely half responded. There were slightly more male respondents (58%). The average age was 24, with a sharp peak around 22 (the author's age). The standard deviation was 159 (to 3 s.f.). Usage
Frequency of use Percentage of respondents 50% 40% 30% 20% 10% 0% several times a day once a day every few days less often once a week

Almost half (45%) of respondents logged into Facebook several times a day. Only 5% of respondents logged into Facebook less often than “every few days”. Note, however, that the invitation to participate in the survey was sent via Facebook, so those who log in less frequently may feel less compelled to respond.

One respondent commented that the frequency of his usage varies dramatically: “Ask me a month ago and I was on several time a day and updating my status all the time. Now I update status once a week! and don't even log on every day.”

A large majority of respondents (86%) logged in for half an hour or less per session. The average reported session duration was 23 minutes, with a median of 15 minutes and a standard deviation of 22.7.
Average session duration (minutes) by frequency
40

30

20

10

Average reported session duration increased as the login frequency decreased, as might be expected, apart from for those users who logged in less often than “every few days”. Perhaps these users do not find the service useful. Indeed, one of them commented: “I signed up but haven't really used it

0 several times a day once a day every few days less often

since. I don't see the point. I use twitter5 to keep in touch with family and I don't really have any friends who use this sort of media.” The survey asked respondents to rate the importance of several key functions of Facebook on scale from 1 to 5 stars, where 1 is labelled “not important” and 5 “very important”, but the meaning of intermediary ratings is not prescribed. The most important reason cited for using Facebook was to keep up to date with their current friends' lives, with an average star rating of 3.9 and the vast majority (89%) giving this reason three stars or higher. It may seem surprising that they would want to communicate via online social networks with people they frequently meet face-to-face and tele-communicate with through other media, but actually it appears that people use social networking services more as a way to view information about their friends' lives that their friends may not communicate to them directly. Indeed, in the case of photographs, where Facebook allows users to easily see photographs in which their friends appear, the media in fact be created by somebody they do not know. It is very unlikely that this type of sharing would happen if it were not facilitated by the social networking platform. In close second and third places respectively were contacting friends (3.7 stars average) and rediscovering old friends (3.6 stars), with around three-quarters (77%) of respondents giving three stars or higher for each. There was no statistically significant difference between the three aforementioned stated reasons. The former suggests that people are using online social networks to replace their address book. This overcomes difficulties with out-of-date contact details that other systems typically suffer from. Contact via social networking platforms perhaps also has different overtones to that of other communication media, especially writing on a user's Facebook “wall” - essentially, writing a message that is intended for that user but can be read by anyone who knows that user. This might be roughly equated to a comment made face-to-face to that person in a public place, where comments will be moderated by other peers. This is often how friendships are formed before any one-to-one meetings are considered appropriate. However, respondents were widely agreed that Facebook was not a place for making new friends, with two-thirds (67%) giving just one star and almost all (93%) giving three stars or less. Men were significantly more likely to give a higher score than women, with an average rating of 1.79, compared to 1.59 for women. It appears that meeting new friends is still an activity that mostly happens offline. However, sites such as HospitalityClub.org6, Meetup.com7 and Match.com8 suggest that many do use the Internet to find new people to meet, when there is some shared interest or motivation. It seems that Facebook is not frequently used in this way, although there are some indications that third-party applications (such as “Are You Interested?”, a simple dating application with around 500,000 daily active users) are starting to fill this void. Number of friends There was a wide variance in the number of friends that respondents reported having on Facebook, from just 2 to 883. The number of responses trails off quickly quickly after 350, with only 10% of respondents reporting having over 350 friends. The average was 212 and the median was 168 with a standard deviation of 159.
5 6 7 8 Twitter is a micro-blogging platform, which can be used via the web or text messaging. http://www.twitter.com A global hospitality-exchange website with around 330,000 members. http://www.hospitalityclub.org A listing of over 74,000 public special-interest meetings across the globe. http://www.meetup.com An online dating website with over 20 million members. http://www.match.com

Several respondents commented on how “friends” is perhaps an inaccurate description of the contacts listed on their Facebook profile, for example: “A lot of the people are on my list because they like to add absolutely everybody they have ever known even if they have no intent of talking to [me], but [they] want to look at what [I] do I guess.” and, more succinctly: “God. So many 'friends' are actually zombie sub-acquaintances!” One respondent mentioned that they thought there was some peer pressure to add as many friends as possible: “Facebook sometimes comes across as a pissing contest to see who has more friends. Lots of people have something like 300 friends, despite studies showing that 100 people is the maximum number of real social interactions you can have9... As a rule, I generally only add people who I see regularly, with some exceptions.” The average number of friends added by users in the last 2 weeks was 3.7, suggesting that the average user's number of friends is growing by 45% per annum, as previous interviews have suggested that they rarely remove friends. The standard deviation was 5.01. There seems to be a significant reluctance to delete friends as it seems slightly confrontational. One user suggested that he didn't like to voluntarily block himself from accessing personal information about somebody: “I know I should delete some of my "friends” but you never know when you might need to creep around their page to see what they have been up to!” The maximum (34) came, unsurprisingly, from the user with the greatest number of friends (883). 20% of respondents stated that they had not added any friends in this period. The survey asked respondents to classify 50 of their Facebook friends into five categories. However, many respondents apparently misunderstood the instruction and classified more than 50 friends. In these cases, it was assumed that they had classified all of their friends. Also, because the list was sorted alphabetically and many respondents (91%) had over 50 friends, the “Family” category may display more bias than the others. Responses with no friends categorised were assumed to have not completed the question.

Friends by category
0.6 0.55

Average number of friends

0.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 Row 73
a close friend you see regularly a friend who was close but you n a new friend who you see regular somebody you don't know well or family Uncategorised

9 This is probably a reference to “Dunbar’s number”, 150, a suggested average size of the social group that is manageable for a human (Dunbar, 2003).

The category into which users placed most of their friends on average was old friends who they do not see regularly, with an average of 18% of their friends. In second place were current friends, with 10%. The other categories had less people in, with an average of 13% in total. Women had significantly more close friends than men, twice as many on average (12% vs. 6%). Respondents only classified an average of 41% of their friends into the suggested categories, suggesting that the categories were not very comprehensive, despite feedback to the contrary when the survey was initially being tested. One respondent, who categorised 62% of his friends, confirmed in a comment that the remainder of his friends don't fit into the given categories. Some respondents suggested additional categories in their comments: “[this question] threw me a bit as there are many people on my first page that I don't really know and don't really meet regularly face to face” “The missing category... is a friend who you contact regularly but don't see regularly. That would be about 25 of the 50 people.” It is also possible that as this question was quite time-consuming and was placed at the end of the questionnaire that some respondents lost patience with classifying their friends. It may also indicate that respondents found it difficult to classify their friends. There were several comments from those who initially tested the questionnaire that indicated that they found this question demanding. News Feed The “News Feed” is Facebook's name for the aggregated view of news by and about all of the user's friends. This view is the first page that users see when logging in, and can be easily reached from any other page by clicking on the Facebook logo. This prominent position within the information architecture implies that Facebook see this feature as a key means of navigation through the service's content. Interest in the aggregated view of all news about the user's friends (the “News Feed”) was very low. The majority of respondents reported finding one-quarter or less of the News Feed items interesting, with just 14% of respondents reporting interest in over half of their News Feed items. It would seem a plausible hypothesis that users would add their closest friends to their online social network first and that a small number of friends should indicate that the user is more selective about who they add. If this were true, users with fewer friends would be interested in a greater proportion of the items in their News Feed. However, there is no significant correlation between the two. Photographs were widely considered the most interesting type of social media, with an average rating of 3.9 stars and the vast majority (92%) giving three stars or more. Friends' intentions to attend events was generally considered interesting, with around half (54%) giving three or four stars. Friend's actions on Facebook were widely considered the least interesting, with almost all (96%) giving three stars or under. Interest in the News Feed was significantly related to interest in profile changes and friends actions (e.g. joining a group, using an application). This suggests that these stories may currently be over-represented in the News Feed, contributing to the low average level of interest.

Interest by friend type
3.5 3.25 3 Average interest rating (1-5) 2.75 2.5 2.25 2 1.75 1.5 1.25 1 0.75 0.5 0.25 0 a close friend you see reguan old friend you don't see or family a new friend you see regularly somebody you don't know well somebody who doesn't fit into any

Respondents reported equal interest in close friends they see regularly and old friends who they don't see regularly, with an average star rating of 3.4 for each. There is no significant difference between any of the four most interesting types of friends. Respondents were significantly more interested in those they don't know well but who they see regularly than those they don't see regularly, with the most respondents choosing 3 stars and 2 stars respectively. This confirms the hypothesis suggested in the initial brainstorming session that finding that user's interest in other Facebook users is influenced by the likelihood of the users meeting off-line. Younger users were significantly more interested in new friends. Almost all users (91%) gave star ratings of 3 or below for those friends who don't fit into the suggested categories, which is significantly lower than all other categories. This confirms that the categories chosen have a strong relation to their level of interest in a given news story. Interest in the News Feed was significantly related to interest in the “uncategorised” people, suggesting that these people are currently over-represented in the News Feed, contributing to the low average level of interest. Key findings Usage of Facebook is generally fast and frequent, similar to email. This does not normally allow enough time to consume all news by all the user's friends, so tools to help the user find the most interesting news are important. Yet the satisfaction with the aggregated news feed provided by Facebook, which is typical of those provided by social networking services, is very low. This confirms that there is an important problem in this area. Keeping up-to-date with current friends and re-connecting with old friends are the two most important reasons for using Facebook, but although they seem like similar tasks, they differ dramatically in the "amount of detail" the user is interested in - the user is likely to want to have their attention drawn to a smaller number of stories about old friend's lives, that are of more significance. Users have almost twice as many "old" friends as "current" friends. Thus displaying a random selection of news will show more from old friends. The proportion of old friends will increase as the user adds new friends, as deleting friends seems to happen rarely (as previously discussed) and research shows that the number of close friends at any one time stays relatively constant, possibly limited by social cognitive abilities (Stiller & Dunbar, 2007).

Interviewees and survey respondents both admitted that many of those on their "friends" list were not people that they knew well. This may be attributed to the apparent social stigma associated with refusing friend requests or deleting friends from the list. Respondents reported being significantly less interested in these people. Interest in the items presented in the News Feed was very low. Survey of published items In order to estimate the number of news stories published by each Facebook user, a small survey was conducted of 20 of the author's Facebook friends, chosen at random. The average rate of publishing amongst these users was 1 story per day. As the average user has 212 friends, this means that there are 212 new stories per day that could be displayed in the News Feed. As the News Feed displays around 20 stories, this suggests that the average user only sees around 10% of the news published by their friends in their News Feed. User perceptions of News Feed Although the News Feed only shows a small proportion of the stories from an average user's friends, there is no opportunity to view the other items or any indication that not all items are displayed. This may create the false impression that all stories from their friends over the covered time period are displayed. The author's Facebook friends were asked to participate in an internet-based survey to ascertain whether users were aware that the News Feed view was incomplete, to which 20 people responded. When asked whether they believed the News Feed to display all stories from their friends or only some, 85% of respondents stated that they believed it only show some of the stories. The participants were then asked how they felt about the News Feed only displaying a fraction of their friends' news. Most felt that it was simply necessary, because the volume of stories is too great: “it feels like a necessary evil, as a news feed of unlimited size would probably be a sensory overload for people with a very large amount of facebook friends.” Some respondents felt that it might mean that they were missing interesting content: “That the boring news Im getting on my news feed doesnt neccessarily mean all my friends are boring.. Im probably just missing all the interesting stuff.” “Disappointed. It's great to see what people are up to.” Many called for filtering mechanisms to allow them to see the most interesting of their friends' news: “it's ok i guess because when i want to see what a certain person does, i visit their page. But i would prefer... to see stories from the people i interact more, either with games, [or] by visiting their page”

“I'd like to keep up with at least the most important news of all of my friends” “I dont care of the life of approximately 75% of my facebook friends so its not really important and I dont want to know everything from my friends' feeds. The good thing would be to be able to select what I want to know from who I want” Utilisation of News Feed filtering features As part of the same survey, participants were asked about their usage of the automatic filtering functions provided by Facebook. Facebook offers two main methods of filtering: ● Feedback buttons: each story displayed in the News Feed has a “thumbs up” icon to give positive feedback and a cross to give negative feedback. This presumably alters which stories are chosen to appear in the News Feed, although no real explanation of how this functions is provided to the user. ● News Feed Preferences: allows users to manually specify which media types to prioritise over others, and also which friends should appear more frequently and which should appear less frequently. Participants were first asked whether they have used the Feedback buttons on the News Feed. Three-quarters of respondents stated that they had not, 10% had used them once or twice and 15% had used them “sometimes”. No respondents stated that they used them regularly. Of those respondents who had not used the feedback buttons, the majority (60%) of respondents said that they hadn't noticed them. This is understandable as the icons are very small and faint. About a quarter (27%) of respondents didn't feel the need to use them and 13% didn't understand their function. Of those that had used the feedback buttons, none felt that they'd had any effect on which items were shown in the News Feed.

The survey then asked the same questions about the News Feed Preferences. Only 35% had ever used them. Of those that hadn't, a large majority (84%) had simply never seen them before. This may be because these feature is accessed via an inconspicuous “Preferences” link at the top of the News Feed. 15% didn't know which items they would find more or less interesting. Of those that had used them, only 27% had noticed an effect. These results clearly show that there are issues with the implementation of Facebook's automatic filtering functions, particularly in terms of findability. One respondent confirmed this in the comments field: “thanks for pointing them out! i'll start using them now :)” Three others discussed how they did not understand the function of these controls:

“i've only used the feedback buttons to block annoying news feed items, i have no idea what effect they have long term” “Don't really know if they just filter out those specific actions, or what really??!” “I understand the concept, but not the execution. Am I saying I'm not interested in photos of that person, or stories about that person, or photos of my friends in general? It's not explained in any detail.” And, more succinctly: “No idea what this button is.” Similarity between Friends The survey attempted to get potential users to classify friends by the type of relationship they share, with only limited success. Another way to look at different friendships is to look at the context in which the develops, such as the workplace, place of study or a club or organisation.

The above graph shows one user's Facebook friends, grouped by co-appearance in photographs together using a force-directed visualisation10. The colouring represents different groups who have many friends in common. It clearly shows grouping by shared contexts, for example the pink represents the members of a University society and the lime-green shows fellow PhD students. In general, those on the right-hand side of the graph are associated with the Computer Science department. This suggests that those who share friendships are also share other similarities, which is supported by Galton (1870), who obtained correlational data showing that in married couples, spouses are similar in many respects. Hammond (2005) argues that these different contexts (which he calls “crowds”) are more meaningful than whether the user has defined them as friends, and that the nature of the relationships with people in a particular “crowd” will often change in the same way over time. Although the Groups function on Facebook could achieve this effect, he considers that instead
10 This graph was created using Touchgraph, http://www.touchgraph.com/

they are generally used more as “badges”; groups of strangers who share a belief or interest. This may be because the Facebook interface does not make it easy for users to browse their friends according to shared group membership, shared friendships, or other similarities. Information about group activity (such as discussions in the group forum) is not even included in the News Feed, so users are likely to even forget which groups they are a member of, unless the admin of the group manually sends them a message or invites them to an event. In turn, the sharing of stories and information is important in bringing people together. Brown & Duguid (2000) argue that rather than the information itself, it is the shared interpretation of these stories that makes individuals feel closer to one another, and that this process of developing a common framework is critical for collaboration around shared information. Although they discuss this in the context of a team of work colleagues, it is easy to see parallels with many social situations, particularly where some group activity is being performed with some goal. For example, sportsmen often tell tales of notable or unusual experiences they've had, often described in specialist language. Other members of their team who are listening to the tale develop a shared understanding of the activity. Similarity, therefore, has implications both for the estimation of social proximity, and hence level of interest in news about that person, and how the information should be presented. Stories from those within a social group will make more sense when told in the context of stories from other members of that group.

Design
I here propose several interface designs for browsing social media in the light of my research findings. Objectives The primary objectives of this part of the project are to: ● explore existing user interface designs for browsing social media and identify their strengths and weaknesses ● to propose a model for predicting the level of interest of social media, within the social context suggested by the research ● to suggest ways in which these model predictions may be usefully incorporated into a social media browser interface that solves some of the problems identified in existing interfaces Design approach Despite the efforts of the research to gain a better understanding of the way people use social networking services and the social media monitoring task in particular, human relationships and social behaviour remains unavoidably complex and hence the the requirements for a social media browsing interface may never be precisely defined. It was therefore considered inappropriate to take a waterfall development approach and instead an iterative development cycle was chosen. For the same reason, it would seem impractical to take a truly human-centred design approach in this context. Also, the research results did not suggest that the user's goals could be rigidly defined. Indeed, it would appear that the users did not see the consumption of social media as a task with particular goals or that they could identify when the task was complete. Rather, the process of gaining social awareness of their community was an ongoing one which they had an intuitive sense was worthwhile. The purpose of a social media news browsing interface is, however, easier to define. Therefore, following arguments made by Norman (2005), it was considered that, armed with our understanding of social context revealed in the research, it would be more productive to now focus on the specific activity of monitoring social media about friends. It is also that the users understand the underlying ideas underpinning the design. This is particularly important in the context of automatic filtering, where “black box” solutions can often undermine user understanding and control. Analysis of social media monitoring activity This research will now focus on one type of information behaviour: the monitoring of the social media activity published by or concerning a user's friends. The overall objective seems to be to gain increased knowledge of the lives of the friends of the user, in order support their relationships. Social networking service users reported that they log into the service regularly, typically daily or several times a day. Several users reported that it was a “reflex” linked with turning their computer on or checking their email. The email checking task may lead naturally into online social network browsing activity as a result of email alerts that draw the user's attention to new content. They typically use a desktop computer for this task, although one interview suggested

that some early-adopters also use mobile internet devices. Users typically browsed “headlines” first - short versions of news items. These include thumbnail versions of recently uploaded photographs or changed profile pictures, notifications of actions such as a friend joining or leaving a group, summaries of profile changes or when a friend signals their intention to attend an event. Users generally reported reading headlines from top to bottom, although several considered that they often “scan” all the available headlines first before reading interesting ones more carefully and perhaps clicking links to view more details about the story. This scanning behaviour is often leads into browsing of other related content, such as clicking on photographs to view the photo album that they are part of, or clicking on an updated profile picture to go that person's profile. Many of the stories that the user finds interesting will lead to further exploration either of the full or related content of the story, or visiting the profile of the author or person it concerns. Sometimes, the user will choose to publicly comment on the content or to send a message directly to the author. This browsing behaviour may continue for some time, until the user decides to return to the scanning behaviour or end their session. Most users interviewed suggested that they usually stop once the remaining items are of little interest to them. The interface designs will seek to improve upon only the monitoring part of the task (viewing the “news feed”), but must make it easy to explore other parts of the social networking site as an integral part of the overall task, and to perform communication tasks. In particular, it must be very easy to view a user's profile. This profile normally has links to the various possible forms of communication. Personas Based on the results of my survey, I constructed three personas that represent three significantly different archetypal Facebook users, based on both their reported activities and motivations. Personas are fictitious, specific, concrete representations of users which provide an effective way to communicate insights gained from user research and to help these insights inform design decisions (Pruitt & Tamara, 2006). The personas were constructed according to the guidelines given by Saffer (2007). Each persona was given was given three types of goals: ● end goals - what they wish to achieve from using the system ● experience goals – how they wish to feel while using the system ● life goals – their longer-term ambitions

Persona 1: Ian - the party animal
Male, 18, student Facebook usage:
● ●

250 Facebook friends logs in almost every day for 20-30 minutes

Interests:
● ● ●

enjoys socialising and meeting new people
photo by Jonathan Melhuish

interested in all aspects of his close friends' lives particularly likes to browse photographs of his friends

Goals:
● ● ●

to know what his friends are doing and what's happening to feel that he's “in the loop” and not missing out on any social events to enjoy his time at University, form close friendships and romantic relationships

Persona 2: Hannah - the chatterbox
Female, 24, translator Facebook usage:
● ●

180 Facebook friends logs in several times a day for 10-15 minutes

Interests:
● ●
photo by Jonathan Melhuish

likes to keep up with all her friends' gossip also interested in occasionally hearing news from school friends

Goals:
●

to know what is happening in her friends lives, so that she can talk about it with them and with common friends to feel emotionally closer to her friends, including those who now live far away to maintain a group of friends to socialise with, to support and be supported by

● ●

Persona 3: Sue – the mother
Characteristics: Female, 32, works in healthcare

●

Facebook usage:
● ●

55 Facebook friends logs in every few days for 10-20 minutes
photo by Jonathan Melhuish

Interests:
● ●

mainly uses Facebook to keep in contact with family in other parts of the world also has some work friends on Facebook, but isn't so interested in their news

Goals:
● ●

to view all of the media produced by and about her family to feel closer to her family, particularly her offspring who have left home in recent years to be a part of her offspring's ongoing development, whilst not wishing to diminish their feeling of independence

●

Analysis of existing interfaces
Analysis of social-networking news interfaces Facebook In order to see how social networking websites support the consumption of social media, I looked at how the “monitoring” task described above would be performed using Facebook, an social networking platform widely used by university students and young people in the UK, USA, Canada and, to a lesser extent, other countries (Burcher, 2008). A key feature of social networking services is that they make it easy for users to see content by or about all their friends, or a particular friend, in a single view. They also make it easy to view more information about this person and to contact them. In this way, the monitoring task forms the “listening” part of an ongoing conversation between a user and their friends. A walkthrough of a typical segment of monitoring activity is performed below using the News Feed view provided by the Facebook social networking platform. The first page displayed after logging in shows the latest news from the user's friends, in the central portion of the page. This is an example of the “center stage” design pattern (Tidwell, 2006). The ordering is not specified, but the newest items tend to be at the top, with similar items being grouped together, such as the updated profile pictures arranged in the grid at the bottom of the screenshot. All the items are fairly recent, perhaps one week old at the most. Each item has a link to the profile of the person it relates to. Some items have links to other content, such as a photo album or video page. Each news item also has positive and negative feedback buttons. When positive feedback is given, the button and the story “headline” is displayed with a green background. When negative feedback is given, the button is shown with a red background and the headline becomes very faint. The user has previously given negative feedback to three stories, and positive feedback to the story at the bottom of the screenshot. As mentioned earlier, the research showed that many Facebook users have not noticed these small, faint buttons. When they were drawn to the attention of one interviewee, she expressed

reservations about clicking them for the first time, as she was unsure of their action. The first time that these buttons are used, a message used to appear informing the user that their feedback will affect whether other users see that story. This did little to re-assure the interviewee, as now she knew that her actions have an unspecified, unseen effect on other users. This message has now been changed to that shown (left), which avoids making any specific promises about what effect it will have, but at least avoids the implication that the actions will have an unseen effect on other users. On this page, the user clicked on the image depicting the video of the “ISKCON Temple in Delhi”. Clicking on the video thumbnail takes the user to the video page. Notice how the blue bars at the top and left-hand side of the screen remain the same, assisting the user in navigating the site by increasing their familiarity with these top-level navigation links. This is an example of the “visual framework” design pattern (Tidwell, 2006). At the top of the page is the profile picture of the author. There are also links to her other videos, videos in which she appears, a link to the main video page, which shows recent videos uploaded by the user's friends. There are navigation links to the next-newest and next-oldest videos. This is an example of the “pyramid” design pattern (Tidwell, 2006). Note that there are no obvious links back to the page from which the user has come, although moving the mouse over the Facebook logo reveals a “home” logo, clicking on which takes the user back to the first page. Perhaps the link is “hidden” in this way in order to encourage deeper exploration of other content. The page has a prominent text field for adding a comment to the item, and a less conspicuous “Share” button that allows the user to add a link to the video from their profile page, or to

send a link to the video page to a friend. There is also an option to “tag” this video with the names of those who appear in it. These names are then shown at the bottom of the video, with hyperlinks to these users profiles. This also allows the video to be highlighted to the friends of these tagged users. The view for viewing photographs is very similar, with the addition of an overview that shows thumbnail versions of all of the photographs in that album or of a particular person. The blue bar along the top is visible on every page, which is an example of the “global navigation” design pattern (Tidwell, 2006). Whilst viewing the video, the user notices that he has received a message, indicated by the “(1)” alongside the “Inbox” link on the blue bar at the top of the page. He clicks this and is taken to his inbox. The Inbox view shows the author's name, the title of the message and the first line of the message body. The message title can be clicked to reveal the full message content. The user instead decides to click the Facebook logo and is returned to the home page. The user scrolls to the bottom of the page, reading each of the items and glancing briefly at each of the photographs. Interviewees suggested that they repeat this monitor-and-explore activity until they judge that the remaining items are of little interest to them.

Analysis of RSS feed-reader interfaces Social networks generally represent a “walled garden” in which users may only share content with other users of that social network. This has many advantages, including easy discovery of content and enforcement of privacy restrictions. However, the use of open standards can lead to greater variety and innovation – for instance, the rise of the World Wide Web at the expense of AOL and Compuserve, both “walled gardens” that were popular in the early days of online services due to their relative simplicity, but ultimately could not compete with the booming

ecosystem of companies and individuals interacting via the open Internet. Although there are significant challenges to be overcome, perhaps a similar pattern will occur in the field of social networking services. It is therefore relevant to look at how the social media monitoring task can be performed using standard protocols and data formats. RSS feed readers help users to monitor web sites by regularly downloading a file from that web site and monitoring it for changes. This RSS file contains several “stories”, which might be text content, such as blog entries, or links to embedded multimedia files, such as photographs, MP3 files, etc. Specialised types of feed readers are available for particular types of feeds, such as “podcatchers” that download linked MP3 files automatically and typically provide facilities for playing them or transferring them to a portable player. When a new item appears in the feed reader, this is indicated to the user. The user can then view the item, or click a link to view the item on the original web site. There is also a link to the front page of that web site. RSS feeds have many different uses and are generally useful for monitoring any online information, but are often used for social media. Almost all blogs offer RSS feeds, and social networking sites also generally offer a selection of RSS feeds, so that users can keep track of activity on the site. This helps to initiate and maintain conversations around social media. However, it is worth noting that it is not nearly as easy to use RSS feeds to monitor news from or about an individual's friends, because discovery of these feeds is difficult, and most people have several feeds, one for each media type that they publish. Similarly, there is no effective way to enforce access restrictions, so RSS is only useful for media that is publicly available. This might explain why RSS feeds have not been widely adopted. This is discussed in more detail in the evaluation section below. In order to identify some common interface features in RSS readers, I looked at the built-in feed reader in the Flock11 and Safari web browsers12, and the Bloglines web-based feed reader13. Please refer Appendix 2 for annotated screenshots of the two interfaces. Flock and Bloglines share the same overall layout, with a list of feeds down the left-hand side of the screen, taking around 20% of the width of the screen and all of the height, with the rest of the area devoted to the feed content. Both show the feed's icon and indicate the number of unseen items alongside the feed name. They also allow the user to re-order the feeds and organise them into folders using drag and drop, although drag and drop is only enabled in Bloglines when “edit” is clicked. In both interfaces, the folders can be “collapsed” to hide their contents. Safari's feed reader takes a different approach, with the feed content on the left, taking around 90% of the display, with only a narrow vertical bar occupying the remaining space on right-most side of the screen. Feeds are chosen using the Bookmarks Bar near the top of the screen, which shows the number of new items since the last time the feed was displayed. Note that unlike the other two systems, this number does not indicate whether the items have actually been displayed on screen, and there is no way to mark items as read or unread. Unlike the other systems, Safari allows the user to filter displayed items by age and sort by title. It also allows the user to view a group of feeds in a single display, with the items displayed interleaved in a single chronology. All three systems display the items in reverse chronological order; newest items at the top. Bloglines and Safari display each story in the full width of the content pane, whereas Flock can
11 Flock, a web browser incorporating facilitatings for publishing and consuming social media. http://www.flock.com 12 Safari, the web browser included with the Apple Macintosh operating system. http://www.apple.com/safari 13 Bloglines, a popular web-based RSS feed reader. http://www.bloglines.com

either display the content in a single-column or two-column layout. In the latter case, the top-tobottom chronological ordering is preserved, with whitespace padding the shorter of the two adjacent stories , at the bottom. Flock uses “infinite scroll”: the scroll bar indicates that the page is not very long, but when the bottom of the page is reached, more stories are loaded so that the page becomes longer. This is presumably to make the scroll bar easier to control when the user is only reading the newest few items, as is normally the case. The user might perceive a “jump” if they are scrolling by clicking-and-dragging the scroll bar as their mouse pointer suddenly changes position, but the transition is smooth if they are scrolling down using the scroll wheel of their mouse. Both Bloglines and Flock mark stories as “read” once they have been displayed, although this behaviour can be disabled in Flock, in favour of clicking a “Viewed” button under each story. Bloglines has a “Keep New” checkbox under each story which will prevent it from being marked as read. Flock instead has a “Save” button which adds it to the “Saved Articles” view. Both interfaces have “Blog” buttons to post the article to a blog, and Bloglines additionally has a facility to send it via email. Flock I then looked at how the monitoring task was performed using the Flock web browser, a desktop application based on the popular Firefox browser that integrates tools for the consumption and production of social media. It has yet to be widely adopted but it shows how social media can be integrated into the browser interface. Below is a walkthrough of a typical monitoring activity performed in the Flock browser. The user opens the “People” sidebar by clicking the icon at the top of the sidebar. The “All” tab is selected by default, which shows content from all the supported social media services that the user has logged into, in chronological order. Only the most recent item from each person (on each service) is displayed. Other tabs show content from a single service. If this tab has unseen items, an orange circle is displayed. Each item has the profile photo of the author, where available. If the item is part of a “media stream”, e.g. photographs, the “Media” icon is highlighted. If the item is text content, e.g. a microblog entry, the first line of the text is displayed. At the top of the sidebar, the icon and name of each service is displayed. If there are unseen items from this source, a blue box is displayed showing the number of unseen items. The green circle indicates that Flock is receiving updates from this source.

The user hovers his mouse over a the first line of a microblog entry to reveal the full content.

The user clicks the “Actions” link to reveal a drop down menu. This menu allows the user to send a direct message to the author, “nudge” him to write a new microblog post, or to view the author's user profile. This is an example of the “extras on demand” design pattern (Tidwell, 2006). The user clicks on “Message”. A new browser tab is opened, with a text field in which the user can compose a message to the author of the content on which he clicked. No login is required as Flock remembers the login details. After clicking “Send”, he closes the browser tab by clicking the “x” symbol on the righthand side of the tab, and returns his attention to the People sidebar.

The user clicks on the “Media” button for anka_gollas, which is highlighted in orange to show that there are unseen items in this feed.

The “Media bar” is opened at the bottom of the browser window. It shows thumbnails of the visual media published by the selected user. The most recent thumbnail is displayed on the left. There are buttons at either end of the horizontal bar to scroll to newer or older items.

The user hovers his mouse pointer over one of the thumbnails to reveal a larger, semitransparent preview of the photograph. Under the photo is the title of the photograph, the name of the author and the author's profile photograph.

The user clicks on the thumbnail and is taken to the relevant photo page on the Flickr.com website. The user is logged in automatically, facilitating easier interaction with all features of the site. On this page comments about the photograph are displayed, and the user can add his own. Hovering the mouse over the image reveals an overlay with buttons to post the image to a blog or to send it via email. There are no further unseen items shown in the People sidebar, indicating that the task is complete.

Conclusions One key advantage of online social networks over RSS feeds for social media is that they make it very easy to find media published both by and about a particular person. In the Facebook News Feed, for example, all media from and about all of a user's friends (and only their friends) is displayed in a single view. On a social networking service, adding new people to this view is easy – simply add them as a friend. With RSS, a feed must typically be discovered by browsing the web. It is difficult to find all the relevant feeds published by a particular person, as most people use different websites for different types of media. Viewing all media produced about a certain person is almost impossible, especially if they have a common name. Crucially, social media within social networking platforms is placed in its social context. Not only is it easy to view all content produced by and about friends, but stories are hyperlinked to the authors profile and the profiles of the people who the media is about (e.g. those who appear in a photo). On these profile pages, it is possible to view all the available media by or about that person, and other key information. It is also possible to see some of the communications that others have had with that person. However, the research indicated that although social networking services effectively aggregate news by and about friends and place it in its social context, the effectiveness of this monitoring interface was undermined by the underlying over-simplistic social model. Users reported that many of their Facebook “friends” were mere acquaintances and that, although many appreciated that there was too much activity to display all of it, they reported very little interest in the news that was shown. Persona-based evaluation of existing interfaces Ian particularly appreciates the “Events” feature of Facebook, which allows him to invited and be invited to events such as parties, concerts, etc., and also informs him which events his friends are attending. He enjoys browsing around the profiles of different people at his University, often friends of friends – a method of exploration facilitated by the online social network. When he identifies somebody who seems to share interests with him, he sometimes writes a message on their “wall” (the public nature of which makes it more socially acceptable to communicate

with strangers), interacts with them via an application (such as sending a virtual “wink” or “kiss”) or takes notes of which events they will be attending. He appears to be more likely to do this when the person in question is female, attractive and single – information that is generally readily available on that individuals profile. However, this exploratory behaviour coupled with his attendance of many social events does mean that the number of friends in his online social network is rising quite rapidly and many of these “friends” are really acquaintances that he has only a passing interest in. Consequently, he doesn't find his News Feed particularly interesting. He has tried using the Preferences to make more news from his close friends appear, but it doesn't seem to have had much effect. Without an easy way to see the news from only his close friends at a glance, he tends to visit their profiles occasionally to check their news, and friends often mention during conversation that they've uploaded or created something so that he will seek it out on. Hannah likes the way that Facebook allows her to see many of the online conversations between her friends (Facebook makes these public via the user's “wall”), and also photographs that they've uploaded of parties or outings together. She finds the News Feed useful for this although she dislikes the way that it appears to show all the news from her friends, but doesn't. Seeing the interactions between her friends' friends makes her feel more aware and closer to her friends, and makes her more comfortable when she meets one of her friends with one of their friends, as she feel she has some sort of relationship with them already; they're not a complete stranger. However, she doesn't tend to view her friends' friends profiles and doesn't contact them. Although she finds it interesting to view news from her school friends and to see how they have changed, it sometimes seems a little irrelevant as she's unlikely to see them again in the near future; re-unions have been discussed but because everyone is now spread over a large distance (often in other countries) it seems unlikely to happen. In order to find the news of her closer friends, she uses the “Recently Updated” list and visits their profiles. However, because this view only shows which fields have been updated rather than any more detail, she finds it difficult to know which updates are most interesting and instead is likely to click through to a profile that she's not viewed recently. In this way, she visits the profiles of around 20 of her closer friends approximately once a week. Sue tends to browse the profiles of her closest family members, rather than trusting the News Feed to display their stories. She is generally interested in any stories that they publish, although she feels as if she is invading their privacy a little when viewing photographs of them at social events, as this is media that she wouldn't normally be shown! If her closest family members don't provide much new content to explore, she sometimes returns to the News Feed to see if there are other stories she is interested in, particularly stories from more distant family members. The News Feed mostly shows news from her work friends, but usually there are one or two stories from distant family members.

Interface concepts
Design 1 - “Traditional” view In the first design sketch, I looked at the key elements of the Facebook “News Feed” and similar interfaces on other sites. This design displays the stories in a chronological order with the most recent items shown first. This design has no “next” button, in common with Facebook, which gives the user a sense of closure and encourages deeper exploration of linked content, but perhaps gives the false impression that it shows all the user's friends' news. This interface has very little opportunity for collecting observed feedback data, apart from when the user clicks to explore the news item in more detail. However, if a user does not click, it does not necessarily mean that they are less interested in that item, as they may simply be satisfied that they have seen enough of the item. In the extreme case, “small” items such as status and profile picture updates are shown in whole in the feed and so no data can be gathered, and there is no representation of read or unread items. Design 2 The second design changes the ordering of the list so that the most interesting items are displayed first, according to some computed estimate. How this might be calculated is discussed in more detail in the “Automatic filtering of social media” section below. The chronological ordering has now been lost, which may be important in some situations, e.g. if one news item is in reaction to another, so an indication of the age of each item has been added. Displaying the age, e.g. “2 days”, was considered more meaningful than simply showing the time and date of creation. This design incorporates buttons for giving explicit feedback to the system, so that it can learn which items the user is most interested in. Hopefully the fact that it is stated openly that the list is ranked by predicted level interest will prompt the users to give feedback to correct the systems mistakes. However, this incentive may still not be enough to provide sufficient feedback to be able to calculate an accurate estimate. This type of layout, although reasonably uncluttered, is not particularly suited to viewing on mobile devices with small screens.

Design 3 This interface design specifically targets mobile internet devices by limiting the display to a single story at a time. Links to user profiles, etc. are maintained, however, allowing for easy exploration of related information. It also aims to increase the amount of user feedback by forcing the user to click a feedback button in order to proceed to the next item. This may initially be seen as slightly annoying and extra cognitive load, but users might appreciate that an accurate estimation of interest can make a big improvement to the user experience when browsing media on a mobile device, where the amount of information that can be viewed “at a glance” is severely limited compared to a desktop display. Design 4

This design radically re-organises the view to be people-centric, showing the stories grouped by who they are by or about. This is supported by the research finding that who the story concerns is a major factor in determining whether or not stories are interesting. It also allows the presentation of stories in the context of stories concerning other similar individuals. As discussed earlier in the “Similarity between Friends” section, similar individuals may share similar stories and presenting them together may help to build a shared interpretation

of social situations, thus helping to support relationships. See the “automatic topography generation for representing social context” section below for technical details of how this could be implemented. Although this could be added to a list view, it would be at the expense of chronological ordering or ranking by predicted level of interest. Moving to a two-dimensional grid allows the representation of both similarity between people displayed and the user's predicted level of interest in them. As similarity is between people rather than stories, and the identity of the author was shown to be the most important factor in determining the level of interest in a story, this view shifts from a story-centric to person-centric layout. Also, as there were significant differences in the level of interest in different media, icons depicting the media type are shown when the mouse is moved over the person's profile picture. Clicking on this icon reveals the stories about this person of this media type. This behaviour forces the user to interact more with the system than the traditional “list of headlines” view. However, this vastly increases the amount of observed feedback the system can capture. Each time the user hovers their mouse pointer over a friend's profile picture, they are expressing an interest in that person. Each time they click on a media type icon, they are expressing a preference for that type of media. As this feedback is received in pairs, there is potential for learning the relationship between friend type, type of media and interest in a story. However, it is difficult to see how explicit feedback could be collected effectively from this interface. There is potential for showing feedback buttons against each story, but the user may be too tired of clicking on other parts of the interface to provide useful levels of feedback on each story. User's profile pictures could be clicked-and-dragged around the grid to give feedback about the level of interest in that user or their perceived level of similarity to others. However, it may be difficult for the user to click and drag across any great distance, thus only allowing easy feedback on those that are only slightly misclassified, which the user may not consider it worthwhile giving feedback on.

Design 5

Design 5 inherits many of the features of Design 4, but instead of organising people into a grid, profile photos are showed in concentric circles, with the inner circle representing the user's closest friends and the position on the circle is chosen in order to maximise each friends' similarity to their neighbours, with each circle also being rotated to maximise the total similarity of neighbouring individuals on adjacent circles. The main advantage of this radial view is that it facilitates easier re-classification of individuals to give explicit feedback to the system about the level of interest in a particular person. Dragging profile photographs between adjacent circles performs a meaningfully large reclassification. Another advantage of this layout is that on average, friends will move between circles less often than they move between rows in the grid view. When a friend moves to a more distant circle, it implies that the user has not maintained contact with this friend and that they are “drifting apart” in their relationship. By highlighting this change to the user, the user may take corrective action, such as viewing their unseen content or making contact. Alternatively, if they disagree with the reclassification (for example, they have communicated regularly outside the social

networking service), they can simply drag the friend's photograph back to a circle closer to the centre. Navigation around the circles is achieved in two ways – by panning and by zooming. Zooming out (using on-screen buttons or the scroll wheel of the mouse) allows the user to get an overview of how many friends have published stories that they have not seen yet, but makes interaction more difficult. Most interaction happens at a “normal” zoom level at which 15-20 friends would be shown at once on an average desktop display. Moving the mouse near the edge of the screen pans the display in that direction. As a result of the topography, moving toward horizontally or vertically changes the predicted level of interest and moving around the circle changes the social context. In this way, it is envisaged that users would generally explore in a roughly spiral fashion, moving in ever-larger circles until they judge that the remaining friends are uninteresting to them. However, if a user currently finds a certain social context particularly interesting, they may explore further out in this direction before returning to closer circles to explore other social contexts. Automatic filtering of social media Several of the proposed interface designs call for some calculated estimate of a story or author's interest to the user. The research findings discussed earlier would seem to suggest that the following factors might usefully predict the users level of interest in a story: ● the type of story (photo, profile change, etc.) ● who the story relates to (close friend, etc.) The research showed that the level of interest in these media and friend types is sometimes affected by simple demographic information; age and gender, which is quick and easy for the user to input. Similarly, the story type may be easily determined from meta-information provided by the online social network. Represent more story types or sub-types than used in this research may improve prediction accuracy. For example, a friend's intention to attend a concert may be more interesting than her intention to attend a poetry reading. Such information is made readily available by the social networking platform. Respondents only succeeded in classifying an average of 41% of their friends into the relationship types suggested by the survey, so it may be possible to devise categories that achieve better coverage. However, respondents considered that these categories represented the most interesting types of friends, which may be what is important in this context. Due to the complexity of human social interaction it may also be justified to use a more complex model of relationship type. For example, K-means clustering could be used to group friends by attributes they share in common, which may usefully allow feedback about one friend to be shared amongst similar friends, increasing prediction accuracy. See also the “Similarity of Friends” section above for a discussion of the sociological basis and the “Automatic topography generation for representing social context” section below for a more detailed description of how friends may be clustered by similarity. If feedback data is available from several users who share friends in common, a collaborative filtering approach may be taken. This has the potential to greatly improve the prediction accuracy, particularly when operating outside of a social networking platform (such as for RSS

feeds), where no data are available to facilitate socially-aware filtering. If, however, these other users are a member of the social networking system, their other visible attributes may be used to cluster them by similarity (as described below), which can provide a good estimate for the initial link weights between the users in the collaborative filtering system. As more feedback is received from the users, the link weights can be adjusted so that they represent their similarity based on their feedback, as in the normal collaborative filtering approach. It seems logical that there would exist a three-way relationship between story type, person type and level of interest. For example, it is more likely that a friend would want to go to a party with a close friend than with their mother. The research approach taken did not permit the collection of data, however it could be collected from real-world usage of some of the abovedescribed interfaces. Also, some story types that are of more significance, such as relationship changes, may be of interest even to old friends who are not interested in the more mundane news of a user. This difference in significance is likely to require a more complex representation than the simple classification employed in this research. The research results did not suggest any straightforward formula that may accurately predict the level of interest in a story using the variables explored in the survey, I propose that a learning approach is required. All the interface designs facilitate the capture of observed feedback concerning interest in an item, to a greater or lesser extent, and all but the first design incorporates a mechanism for the user to provide explicit feedback. Once observed and explicit feedback has been collected, we must devise a way to relate this feedback to the attributes of the story (e.g. relationship with the author, story type). There are a multitude of machine learning techniques that could be applied to this problem, including neural networks and other supervised learning techniques. We can see this as a dimensionality reduction problem, as we have a high-dimension input space of attributes that we wish to map to a one-dimensional value. A commonly used method is Principal Component Analysis (PCA), which calculates the eigenvalue decomposition of a data covariance matrix to find the most “descriptive” projection of the high-dimensional data onto a low-dimensional plane (Shelns, 2005). PCA would have to be run regularly, perhaps daily, which may be computationally expensive. An alternative would be to use an online learning system, which can continually adjust the significance of each contributing factor, such as an Artificial Neural Network. It is important, however, that even explicit feedback does not have too strong an immediate effect in order to allow the calculation to take account of data collected over a longer period of time, as an individual's estimation of another may, for example, be significantly affected by their present mood (Berry & Hansen 1996). Automatic topography generation for representing social context Design 4 and Design 5 above call for a method of automatically arranging individuals so that similar individuals are shown together. There are several factors generally available on social networking platforms that might sensibly be incorporated into this estimation, such as: ● shared friendships (where two individuals are friends with the same third party) ● co-appearance in photographs (where two individuals have been tagged in the same image) ● public communications between two individuals (e.g. “wall” posts) ● shared group membership, university course attendance, employer, residential address ● shared interests, activities

As there is no apparent way to simply combine these values, they represent a hyperdimensional data space upon which we must use a dimensionality reduction technique in order to fit individuals within the constrained 2D topographies proposed in the interface designs. The interface designs call for sorting by predicted level of interest in one dimension and similarity in the other. The level of interest must first be reduced to a single value as described above, which can than be used to rank the individuals and sort them into the correct size groups. In Design 4, these groups are of fixed size according to the width of the display, in Design 5 the group sizes increase when moving outwards from the centre of the display. In both, there exist sufficient groups to accommodate all individuals with unseen news. Each of these groups must then be arranged so that neighbouring individuals are similar according to the attributes suggested above. Self Organising Maps are a form of Artificial Neural Network that use competitive learning to create a low-dimensional representation of a real-valued hyperdimensional space that preserves a notion of how the cluster centres relate to each other, by linking them together in a grid as depicted left (Crnkovic-Dodig, 2007). Although this is a 2D image, the data points (shown in black) may exist in hyperdimensional space while the grid exists in a lower number of dimensions. In the case of Design 4, we use one dimension (a line) and in the case of Design 5 we use two dimensions (a circle). A Kohonen network uses the following approach to achieve this dimensionality reduction whilst preserving structure: ● a data point is selected at random ● the closest cluster centre (also referred to as codebook vector) to the data point is found ● this cluster centre is moved towards this point in the hyperdimensional data space by some fixed amount, the learning rate ● all other cluster centres are also moved toward the data point, but by a smaller amount – at each link in the grid, the amount that the cluster centre is “dragged” toward the data point is reduced, usually according the a Gaussian function The latter step ensures that some structure is preserved, thus ensuring a meaningful relationship between the cluster centres. In the case of our social media display, this ensures that those who appear close together within the topography are somehow similar, based on the attributes suggested earlier. Thus, the structure of the Kohonen network should mirror the on-screen display. Although the structure of the grid may be trivially adapted to virtually any shape, such as the circular layout called for by Design 5, it is possible that this may not generate optimal results, because as the circle is deformed into an ellipse, cluster centres that are far apart on the circle may become neighbours in the data space. One possible solution may be cyclic maps as employed by Lohninger (2004), but this requires further investigation. The final stage is to rotate the clustered groups such that the similarity between adjacent individuals across row or circle boundaries is maximised. As each group has been arranged independently to maximise the similarity of neighbours only within that group, it is unlikely that a perfect match between rows or circles will be achieved. This correlation could only be improved at the expense of the ordering by predicted level of interest. If it were decided that similarity between rows or circles is more important, the method could be adapted.

Prototype implementation
Specification It was decided to implement a digital prototype of Design 5, in order to gain user feedback. The prototype should demonstrate the key elements of the interface described above and realistically create the impression of operating within the context of a social networking service. The prototype will be evaluated qualitatively through an online questionnaire. Technical details The interviews indicated that most users use a web browser on a personal computer to browse social media, although one early-adopter also used a mobile internet device (iPhone) equipped with a fully-featured web browser (Safari). For this reason, the prototype will be implemented as a web application, with some consideration given to use on smaller devices. Despite the author having very little previous experience with the Ruby language, it was selected due to its flexibility and because of the availability of the Rails web application framework, which allows the rapid development of web applications. It is particularly suited to the iterative development approach taken in this project. The application was developed on an Apple Macintosh, which offers a UNIX-based operating system for which a wide variety of software is available and is well-suited to program development. The Mongrel web server was used with Locomotive, a pre-built binary package that allows Rails applications to be quickly deployed within a standard, fully-functional Rails server environment. System architecture The prototype system was designed using the Model-View-Controller (MVC) architectural pattern. This pattern seeks to separate the logic parts of the program from those that affect how the output is displayed, allowing one to be easily changed without affecting the other. The functions of each of the three sections are briefly outlined below (Freeman et al, 2004): ● model: holds all of the data, state and application logic. In essence, the model represents real entities such as people or news stories. It offers methods for getting and modifying its state (and validation), but nothing else. ● view: controls the presentation of the data, for example the HTML, CSS and Javascript required to format the data as a web page. ● controller: provides the interaction between the other two. This is where most of the application-specific code is. For example, clicking a button in the view causes code to be executed in the controller which defines the action taken, and interacts with the relevant model(s). Ruby on Rails facilitates developers in using the MVC pattern through its default behaviours and the structure of the standard scaffold code. For example, models automatically interface with their corresponding data table and controller methods are automatically executed when the corresponding URL is requested. Rails also makes it easy to use RESTful interaction, a pattern which was applied in this project. The statelessness constraint is achieved using the standard Rails session facility, which sets a cookie in the browser with a unique session ID but stores other session data in a database table

on the server, for security reasons. The radial view (based on Design 5) uses CSS for the layout and Javascript to provide interactivity. The JQuery Javascript library was used to provide the pop-up overlay window in order to speed development. Additional custom Javascript code was added to control the hoverover behaviour. Models – database tables Model name News Person Person_type Story Story_type Attributes Links to Person and Story, URL of the story content, time the story was created First name, Surname, Profile photograph URL, link to story type Label (text description) Story headline Label (text description), default content URL

In retrospect, it was a mistake to have chosen an uncountable noun as a model name (News). Rails automatically maps singular model names to plural controller and view names. This helps clarify whether the object represents a single entity or whether it is working on a collection of objects of that type, but only when using a countable noun. It is also slightly confusing to speak of creating a “new News object”!

Models – methods All models automatically have getters and setters methods for of the above-mentioned attributes, provided by ActiveRecord, which is part of the Rails framework. Classes not mentioned below have no additional methods. Model name Method name News makenews! generate(person) Function Iterate through all people, call generate with probability 0.5 Generate a news story about person by creating a link between the person and a randomly-selected story Return the URL of the news item relative to the web root (“/”) Return the first name and surname, concatenated Return true if there are any stories of story type “name” Return a string containing the name concatenated with the story headline for the first occurring news story of the given type Return an array of Story objects matching the given type Return an array of News objects matching the given story type

link Person name has_stories?(name) headlines_by_type(name)

stories_by_type(name) news_by_type(name)

Models – validation Model Person Story StoryType Field firstname surname content content name name Validation Length is between 2 and 100 characters Length is between 2 and 100 characters Length is between 8 and 100 characters Uniqueness Length is between 2 and 100 characters Uniqueness

This validation is by no means comprehensive and should be expanded if this prototype is developed further. View View name list show radial Controller News People People Function List all news Show details of a single person, display all stories relating to them List all people and all news about each person

Controllers The Rails scaffold code generated for controllers includes methods to: ● list all instances of the corresponding model ● edit, update or delete, show a particular instance of the corresponding model ● create a new instance of the corresponding model These methods are omitted below except where modified. The People controller has only these methods and is hence omitted. Controller name Method name News list radial Function return all news items return all people and array of positions in which they should be displayed

For the purposes of the prototype, the list and radial methods of News call News.makenews! in order to generate new news each time the view is generated. Testing Automated testing can help to support iterative development by provide quickly highlighting problems, thus giving developers confidence to constantly make changes to the code without worrying about unforeseen knock-on effects on other parts of the program. Some developers even write the unit tests before the code they test, thus creating a programmatic development roadmap that provides instant and unambiguous feedback on progress (Larman, 2004). Unit tests are small sections of program code that perform an action (such as calling a method) and assert a certain outcome. If the actual outcome causes the assertion to be false, the programmer is alerted. The following unit tests were written: Model Test name News test_items_are_generated

Description Call method to generate new story, check that there is one more story afterwards Try to create and save a Person object with no firstname, check that it creates the relevant error upon trying to save Try to create and save a Person object with no surname, check that it creates the relevant error upon trying to save

Person test_blank_firstname

test_blank_surname

test_creating_invalid_firstname Try to create and save a Person object with a firstname of a single character (too short), check that it creates the relevant error upon trying to save test_creating_invalid_surname Try to create and save a Person object with a surname of a single character (too short), check that it creates the relevant error upon trying to save

These tests do not represent a comprehensive test set and should be expanded before further development of this prototype.

In addition to automated testing, manual browser-based testing was conducted regularly throughout development. The iterative development approach taken facilitates this testing, as it emphasises producing incremental improvements and producing a working version frequently, which can easily be tested manually. It would be difficult to automate testing of the layout and client-side interactive elements, although a certain level of automation could be achieved using the Selenium browser-based automated testing tool, for example to assert that a certain piece of text is present on the page or that an error message is not displayed.

Prototype Evaluation
Walkthrough of typical monitoring task Upon loading the digital prototype, the user is presented with a view depicting his closest friends on his social networking service who have published news stories that he has not viewed previously. Those in the innermost circle represent those who the system predicts he is most interested in. The research shows that this would be a big improvement for many users, by allowing them to prioritise their closest friends, without the system hiding anything from them. This allows users to view their close friends' profiles more often, but allows them to explore others occasionally. This behaviour is also supported by the socially-aware topography. The position around the circle represents the social context, with similar individuals placed close together. If a user currently finds a particular social group more interesting than usual, he may explore further in this direction. The user moves his mouse pointer over one of the profile photographs, revealing icons that represent the different media types present in that friends' unseen stories. There may be several stories of the same type. By using “drill-down” structure, news from lots of friends can be represented without a cluttered display. More importantly, this high level of interaction generates far more observed feedback for the system to learn from.

The user clicks on the icon representing a photograph, opening the relevant story in an overlay window. By removing the need to reload the whole page in order to view a story, exploration of news items is accelerated and the abrupt transition to a new page is avoided, encouraging users to explore more items. This click also provides valuable observed feedback for the system to learn from, not only indicating a high level of interest in this friend but also indicating an interest in this media type. Conversely, if the user does not click, it can be assumed that they are interested in this friend but not in the media types offered. This story view could have explicit feedback buttons for each story, but the success of the interest prediction is not reliant on their use. After exploring several items from his closest friends, the user moves his mouse pointer to the edge of the screen and the view pans smoothly in this direction - the view has “momentum” in order to pan more smoothly. By moving his mouse pointer to a position on the radius that lies between Lara and Marianne, he indicates that he wishes to see the friends that are most similar to these two individuals; in this case, friends from his photography class. Other members of his photography class are displayed. After exploring some of their news items, he decides that rather than continue in the same direction towards people he is lessclose friends with, he will pan around this circle towards people from another social context.

The user decides to click on “Lucy” taking him to that user's profile page. The profile page has links to all the stories by or about this person, a profile photograph and other key details. This view is not the focus of this research and so alternative designs were not explored.

After panning almost all the way around this circle, the user zooms out (using the scroll wheel on his mouse) to view all of his friends at a glance. He decides not to explore the media of the remaining friends and considers the task complete. Unlike the Facebook News Feed, he knows that this interface shows him all of his friends with unseen media, and that he is then able to choose which to ignore, rather than this being decided by some hidden algorithm that is not explained.

Persona-based evaluation Ian uses this view mostly to monitor the activity of his closest friends, in the innermost 2 circles. Despite having a large number of Facebook friends, he rarely explores the media of those in the outer circles, although he does admit that it looks quite cool to zoom out and see all the current photos of the friends he's “collected” - this view changes as his friends update their profile pictures. However, he suggests that he might use it more if the display could somehow show the friends of his friends, perhaps those that he's predicted to be most interested in, so that he could use the interface for also exploring other profiles. Hannah explores a little more widely, as she's interested in the activities of a larger group of friends; she always explores the media of those friends in the first two circles and explores the media of those in the third circle every few days. She rarely explores the outermost circles, however. As she quite often explores the media of many people in one session (often over 20), she sometimes gets slightly tired of the repetitive clicking. However, she appreciates that these actions have some effect, and particularly likes the way that friends visibly “drift away”. She often contacts these friends as a result, and likes the way that this feature helps her to actively maintain her relationships. Sue's view almost represents a circular family tree, which made her slightly surprised when she

first encountered a colleague in one of the outer circles! She finds the view intuitive and likes the way that her closest family are in the centre. However, she doesn't like the way that it only shows those people who have unseen media – she finds it disconcerting to log in and find that some of her close family have disappeared. She would prefer that they always stayed there but were somehow highlighted if they had new stories. As she has relatively few friends in her online social network and many of them are not very active, the display is often quite small, with only two circles appearing. User evaluation Explain method A screencast video was prepared showing the key elements of the prototype interface in action and explaining the significance of the layout. A copy of the video is included on the accompanying CD and is also available online at http://youtube.com/watch?v=4AbWpW-wuuk The Facebook friends of the author were invited to participate in an online survey where they were shown this video and then asked:
● ● ● ●

What do you think of the "circles" design? Do you think you would like to use it? Is it better than the current Facebook News Feed? Can you see any potential problems?

Out of the 148 people invited to participate in the survey, 11 responded (7.4% response rate). Three respondents expressed reservations about the amount of interaction required: “What I like about the news feed is that it is push news, rather than pull. I can casually hop on to Facebook, see what people are doing, and go back to whatever I was doing. If I had to deliberately choose specific people to click on to see any news, I probably wouldn't bother - if I was interested in that person's news I'd look at their profile page, mini-feed and wall.” “Quite a nice design, well done! Yet, not sure if I will use it, I find it quite boring to have to click on photos of everybody you're interested in to see their news. Actually instead of replacing the current News Feed, I think that your idea could complement it for those of us who want more detailed information on some specific people.” “I think it could be quicker and easier to scan through a list of news feeds on a page, rather than having to actively click on people one at a time.” The concept of organising friends by estimated social proximity was widely welcomed: “Having a lot of Facebook friends, I wish there was a better alternative to being able to manually add a limited number of people to the list of those you do and do not wish to see news about. Many of the people on my list are friends from school who I don't need to know about day to day, and I'd like to be able to tell Facebook that.”

“I think it would make navigating through news stories a bit easier. I would definately be interested in using a system in which I could look at news feeds of the specific people I am interested at the time.” “Being able to drag and drop, and having people drift naturally, is a great idea.” “the fact that the computer could choose to show you news feed from your most clicked profiles is definitely a good idea” One user disliked the idea of removing some of the serendipity of the News Feed: “narrowing down the news feed could take away some of the surprising to find out something for someone that you don't expect.” One respondent raised concerns that they might miss interesting stories from people they're not normally interested in, although as they realised that the Facebook News Feed didn't even show them all the news from their friends, they were presumably not suggesting that the proposed interface was necessarily worse in that respect: “a more distant friend who you would probably never actively click on might occassionally publish some really interesting news, which you would then miss.” There was significant disagreement about how the design would function with different numbers of friends displayed, with some respondents believing that it would be of most benefit to those with large numbers of friends but others feeling that the display would become ineffective when large numbers of friends needed to be displayed: “out of curiosity why a circle? surely a circle is fairly inefficient spatially, can you imagine what it would look like with 200 friends say?” “I don't think the circle design would work very well because I don't think it would be able to fit all my friends on facebook into it!” “the circles idea is pretty interesting, althought I'm not sure if it would work with people who have lots of friends” “The circles idea would work for people with large numbers of friends, so this idea isn't aimed at the way I use facebook.” It is possible that this confusion was caused by the use of too few friends in the prototype, which meant that all of the friends could actually be displayed quite comfortably on a desktop display. This view was the first that was shown in the video, perhaps giving the respondents the false impression that all of their friends would be displayed at once. This explains the first comment, which presumably refers to the large gap between the outer circle and the rectangular display when showing all friends. Similarly the second and third comments indicate that they assume the number of friends displayed is either fixed or that the size of the profile photographs would be decreased in order to fit more friends onto the screen, thus making it difficult to use.

Justifiably, one user had reservations about how well this would work on mobile devices or for blind people: “I don't see how it would work properly on anything other than a PC with a decent size screen and the latest browser (i.e. PDA, mobile phone, text-to-speech).” Blind users are likely to be served much better by a list-based view, which should be offered as an alternative. Determining usability on mobile devices requires further research; for example, it may be found to be quite usable on the Apple iPhone, which has a relatively large screen and a multitouch interface that allows for natural panning and scrolling. However, the icons representing the media types would undoubtedly require enlarging significantly.

Conclusion
The fundamental advantages of social networking services stem from their electronic representation of social structure. For example, the problem of displaying socially-relevant digital media is effectively achieved by displaying to the user content that is either published by or concerns their friends. This research has shown that people are using social networking services to support a wide variety of relationships. The ability of the social networking service to display interesting news is being undermined by the over-simplistic underlying social model. The answer does not seem to be to ask users to manually classify their friends. Although respondents considered that the proposed categories described the most interesting individuals in their online social network, on average they only managed to classify a minority of their friends, giving little clue as to their relationship with the remaining others. Additionally, this task was found to be demanding and time consuming. Instead, it is proposed that user interfaces may be devised that draw on machine learning techniques to assist the user in filtering and prioritising social media. These machine learning techniques allow a far more complex and descriptive social model to be constructed, potentially with little extra effort on the user's part. Several interface designs were proposed, one of which was developed into a digital prototype which elicited positive initial reactions from social network users. Further Work Although the feedback received on the digital prototype was generally positive, some respondents expressed concerns about the amount of interaction required and how the design would adapt to accommodate a larger number of friends. Further work is required to produce an online prototype that users can interact with, and with a larger number of friends displayed. Also, users' first impressions may not reliably indicate their long-term behaviour, so ideally the system should be extended to display real data gathered from a social networking service and subjects should be asked to use the system over a longer period of time with real data, as part of their normal monitoring activity. Sadly Facebook makes development of this type of application difficult, but perhaps the system could be extended to interact with another social networking platform, such as one that supports the OpenSocial14 standard. The implementation of a system that uses real data from a social networking service would require the application of the machine learning techniques described. Although the techniques proposed seem suited to the task, their performance has not been tested. The proposed approach should be tested on real data and the techniques optimised iteratively. There are many different topologies that may be devised; a grid and concentric circles are presented here but are not necessarily the optimal topologies in terms of their ability to describe social similarity of and the level of interest in the individuals displayed, and the usability of the interface. The former may be explored empirically based on collected training data and data from the social networking service. The latter would need to be evaluated through user trials.

14 OpenSocial defines a common API for social applications across multiple websites: http://code.google.com/apis/opensocial/

References
Berry, D.S. and Hansen, J.S (1996). “Positive affect, negative affect, and social interaction”, Journal of Personality and Social Psychology, 71, pp. 796-809. Brown, J. and Duguid, P. (2000), The Social Life of Information, Harvard Business School Press. Burcher, N., (2008) “Latest Facebook usage statistics - users, networks, groups, applications and country stats”, http://www.nickburcher.com/2007/12/ latest-facebook-usage-statisticsusers.html [accessed on 3rd April 2008] Crnkovic-Dodig, L. (2007), “The Self-Organized Gene (Part 1)”, http://blog.peltarion.com/2007/04/10/the-self-organized-gene-part-1/ [accessed 2nd April 2008] Dunbar, R. (2003), The Tipping Point, Wheeler Publishing. Dunbar, R. and Spoors, M. (1995), “Volume 29, Issue 1, January 2007, Pages 93-104”, Human Nature, Volume 6, Number 3 (1995) pp. 273-290. Freeman, E., Sierra, K. and Bates, B. (2004), Head First Design Patterns, O'Reilly. Galton, F. (1870), “Hereditary genius: An inquiry into its laws and consequences”. [online at http://galton.org/books/hereditary-genius/, accessed 26th March 2008] Garton, L et al, “Studying Online Social Networks”, http://jcmc.indiana.edu/vol3/issue1/garton.html [accessed 3rd April 2008] Groh, G., Ehmig, C. “Recommendations in Taste Related Domains: Collaborative Filtering vs. Social Filtering”, Proceedings of the 2007 international ACM conference on Supporting group work, Sanibel Island, Florida, USA, pp. 127-136. Hammond, S., “Crowding out the Social Graph”, http://simonhammond.com/blog/2008/03/08/crowding-out-the-social-graph/ [accessed 23rd March 2008] Jackson, R. M. (1977), “Social structure and process in friendship choice”, in Fischer et al. (eds), Networks and Places: Social Relations in the Urban Setting, Free Press, NY. 59--78. Larman, C. (2004), Agile and Iterative Development: A Manager's Guide, Addison-Wesley. Lenhart A. and S. Fox (2006), “Bloggers: A portrait of the internet's new storytellers”, Pew Internet and American Life Project [online at http://www.pewinternet.org/PPF/r/186/report_display.asp ] Lin, P. and McDonald, D. (2006) “Exploring the social realities of online communities through the lens of a Human Information Behavior framework”, The Information School, University of Washington. [online at http://dlist.sir.arizona.edu/1521/, accessed 28th January 2008]. Lohninger, H. (2004), “Kohonen Network - Background Information”, http://www.lohninger.com/helpcsuite/kohonen_network_-_background_information.htm [accessed 2nd April 2008]

Morita, M. and Shinoda, Y. (1994), “Information filtering based on user behavior analysis and best match text retrieval”, Proceedings of the 17th annual international ACM SIGIR conference on Research and development in information retrieval, Dublin: Annual ACM Conference on Research and Development in Information Retrieval, pp. 272-281. Norman, D. (2005), “Human-centered design considered harmful”, interactions 12(4):14-19. Parks, M. and Roberts, D. (1998), “Making Moosic: The Development of Personal Relationships on Line and a Comparison to their Off-Line Counterparts”, Journal of Social and Personal Relationships, Vol. 15, No. 4., pp. 517-537. Pruitt, J. and Tamara, A. (2006), The Persona Lifecycle: Keeping People in Mind Throughout Product Design, Morgan Kaufmann. Saffer, D. (2007), Designing for Interaction: Creating Smart Applications and Clever Devices, New Riders. Shelns, J. (2005), “A Tutorial on Principal Component Analysis”, http://www.cs.cmu.edu/~elaw/papers/pca.pdf [accessed 2nd April 2008] Sonnenwald, Diane H. “Evolving perspectives of human information behavior: Contexts, situations, social networks and information horizons”, http://dspace.bib.hb.se:8080/handle/2320/2209 [Accessed on 3rd April 2008] Stiller, J. and Dunbar, R.I.M. (2007), “Perspective-taking and memory capacity predict social network size” in Social Networks, Volume 29, Issue 1, January 2007, Pages 93-104. Tidwell, J. (2006), Designing Interfaces, O'Reilly. Toffler, A. (1970), Future Shock, Random House. Wikipedia (2008), “Social Media”, http://en.wikipedia.org/wiki/Social_media [accessed on 3rd April 2008] Zajonc, R. B. (1968), “Attitudinal Effects of Mere Exposure”, Journal of Personality and Social Psychology, 9, 2, 1-27. Zhou, W., Sornette, D., Hill, R. and Dunbar, R. (2005), “Discrete hierarchical organization of social group sizes”, Proceedings of the Royal Society B: Biological Sciences, February 22; 272(1561), pp. 439–444.

Appendix: CD structure
Directory name interviews user research prototype code screencast Contents Notes and recordings from initial interviews Survey questions, responses and statistical analysis for both surveys The Rails web application code for the prototype application A copy of the screencast video used in the user evaluation of the digital prototype

Please note that a feature of Ruby on Rails that facilitates the rapid development of web applications is the generation of standard “scaffold” code and configuration files. Unless specifically commented otherwise, it should be assumed that the code was not written by the author. Please refer to the Prototype Implementation for details of files that contain code written by the author. To deploy the prototype code, simply copy the Rails application code into the relevant directory of a Rails server. It has been tested using Mongrel within Locomotive, but should work in any Rails server with no modification. The application can be viewed in any browser but has been tested most extensively in Opera.