Collaborative Visualization of Large Scale Hypermedia Databases (29-Mar-1996)
Page 1 of 7
Retrieve this document as Postscript (195K, Compressed with gzip)
Collaborative Visualization of Large Scale Hypermedia Databases
Chris Brown, Steve Benford and Dave Snowdon Communications Research Group, University of Nottingham, University Park, Nottingham NG7 2RD ENGLAND Tel: +44 (115) 951 4226 E-mail: {ccb, sdb, dns}@cs.nott.ac.uk
Table of Contents
1. Introduction 2. Techniques 3. Implementation 4. A specific application: The Internet Foyer 5. References
Abstract
We begin by reviewing techniques for visualizing large scale hypermedia databases. We present a definition of large scale databases, introduce a scoping technique to handle them, and discuss collaboration support. This leads to a discussion of the implementation; we discuss browsing and searching, and the embodiment of database users in the visualization. Finally, we present an example application of these techniques: The Internet Foyer.
1. Introduction
This paper proposes techniques for visualising large-scale hypermedia systems such as the World Wide Web and then populating the resulting visualisations with many mutually aware, co-present and communicating users. It also describes an initial implementation of these techniques. There have been two general motivations for this work:
l current 2-D interfaces to hypermedia systems typically display only one page at a time and
fail to indicate the location of this page within its broader context. This leads to disorientation and users becoming lost.
l although such systems are multi-user, current interfaces ignore the presence of other users.
In other words, users remain invisible to one another and are provided with no opportunities for mutual interaction and cooperation.
file://I:\GMD-paper.html
3/8/00
Collaborative Visualization of Large Scale Hypermedia Databases (29-Mar-1996)
Page 2 of 7
The more specific aims of our research have been to develop collaborative visualisation techniques to: handle large scale hypermedia databases; view and manipulate objects in these databases; find information held in these databases; have an awareness of other people accessing these databases; and find people with whom you share interests who are also using these databases. The key challenge posed by distributed hypermedia systems is that they are large scale. A large scale database is defined in this paper as a database which is too large to display and search in its entirety, in real time. Thus the qualifying size of a large scale database depends on a function of the computational power available to process it. Our approach has been to exploit the emerging technology of Collaborative Virtual Environments in order to construct a range of three dimensional hypermedia visualisations to support browsing and searching activities. These visualisations are inherently shareable by graphically embodied users who are able to communicate with one another over audio and other channels. In addition, we consider the representation within these visualisations of the presence of other ‘2-D’ users (i.e. ones browsing through currently available interfaces). To address the problem of scalability, we propose a general framework of scoping mechanisms which offer users powerful support for expressing their current interests in information and which allow the underlying system sufficient flexibility in interpreting these interests. Finally, as an additional twist, we consider how such technology can be deployed in the form of publicly shareable projected displays - the Internet Foyer - adding a further dimension to the idea of collaboration within and around large scale hypermedia systems.
2. Techniques
We structure our discussion of underlying techniques into three parts: visualisation styles, scoping and collaboration support.
2.1 Visualisation styles
We propose that several different visualisation styles will be required in order to successful support all of the functionality of hypermedia systems. In particular, we consider the need to provide visualisations to support both browsing and searching. Browsing requires the construction of a map interface to display an overview of many linked hypermedia pages. Two approaches to this problem are: constructing hierarchical representations of the underlying WWW network structure using spanning-tree techniques and rendering these using ‘cone tree’ [Card et.al. (1991)] techniques; and constructing a 3-D connected graph representation using the Force Directed Placement technique [Fruchterman and Reingold (1991)]. These interfaces also need to support other functionality that is traditionally found in hypermedia browsers such as the ability to read the contents of individual nodes and to show a history of nodes which have been recently visited. Searching requires the construction of interfaces that can dynamically display similarities between objects calculated from users queries as visual characteristics, especially proximity. A number of such techniques exist including the simulated annealing techniques applied by Matthew Chalmers in the BEAD system [Chalmers and Chitson (1992)] and the multiple search query technique first proposed by Olsen et.al. in the VIBE system [Olsen et.al. (1993)]. A key aspect of such interfaces is that they could allow users to dynamically interact with search queries and results in a way that is not possible with traditional single query text-based retrieval systems as are currently found on the WWW. file://I:\GMD-paper.html 3/8/00
Collaborative Visualization of Large Scale Hypermedia Databases (29-Mar-1996)
Page 3 of 7
2.2 Scoping
We argue that a prerequisite for successfully visualizing large scale hypermedia databases is the ability to control the number of objects in the visualization so as to be within the real-time constraints of the host computer. We define scoping to be the technique of constructing a manageable subset of the overall hypermedia database which reflects a combination of a user’s interests and underlying system constraints. Considering the problem of scoping in more detail, the real time constraints of a hypertext database rendering system can be described in terms of O (objects which can be rendered in real time) and r Oc (objects which can be computed in real time). If O > Or, rendering capacity is the limiting c factor and traditional visualization-based scoping mechanisms such as distance-based level of detail control, spatial partitioning or spatial awareness levels can be used. However, if O > Oc, a scoping r mechanism which reduces the number of objects to be computed or their complexity in each iteration needs to be used. We further propose that, in order to more fully understand the complexities of scoping, it is necessary to consider a hypermedia system as consisting of three distinct layers. This layering approach grants a level of separation between the visualization of the database, the structure of the information itself (Logical Layer), and the inner workings of the particular implementation of the underlying database (Organisational Layer). This layered approach is shown in Figure 1.
FIGURE 1. Hypertext database architecture, as layers We propose that different scoping mechanisms can be applied to each layer of this architecture. We argue that through a combination of these parameters at different database levels, users will be able to scope information according to a precise specification of their interests combined with sensible system rendering and searching constraints. A breakdown of some possible mechanisms into the three level architecture is as follows:
Filter Location Cost Time of Day Protocol Bandwidth Access Permissions Example Server name, server location and position of object in server Cost of accessing the object When scoping should be applied Only access objects via ftp or some other protocol Only access objects available at > 50K/sec Whether access permission is granted or not
Timeouts & Server Errors How to respond to various errors
TABLE 1. Organisational level scoping parameters
Filter Link distance Based Link type on Structure Number of incoming links Number of outgoing links Example Only include objects within N links of a central object Only follow links of a certain type (where supported) Only include objects pointed to by at least N others Only include objects with links to at least N others
file://I:\GMD-paper.html
3/8/00
Collaborative Visualization of Large Scale Hypermedia Databases (29-Mar-1996)
Page 4 of 7
Object Size Object Type Based Object Header Info Object Title on attributes Object Keywords Object Expiry date/time Body Data
Only include objects larger than X Only include HTML, VRML etc. objects Only include based on other arbitrary header information Only include if title matches that specified Predefined or algorithmicallyy generated Only include objects which have not yet expired Only include objects containing the word "X"
TABLE 2. Logical level scoping parameters.
Filter Example
Proximity of object Proximity to the user of the visualization Number of objects Only display a maximum of N objects Frame rate Cull objects until acceptable frame rate is achieved
TABLE 3. Visualization level scoping parameters.
2.3 Collaboration support
Next we consider the addition of various collaboration support mechanisms to these visualisations. In order to enter into collaboration, users must first have an awareness of each others presence and activities. Such awareness might be provided through embodiments and representations. Embodiments refer to graphical portrayals of other users who share the same visualisation with oneself (see [Benford et.al. (1995)] for a general discussion of embodiments). Thus, many mutually embodied users may share a common visualisation, each seeing it from their own viewpoint in 3-D virtual space. Representation, on the other hand, refers to graphical portrayals of users who are accessing the information in ones visualisation by some other kind of interface and so who don’t share a common virtual world with oneself (e.g. showing representations of Mosaic or Netscape users as the wander across representations of WWW pages in some 3-D visualisation. Once mutually aware, users might enter into direct communication, either in real time through live audio or text channels or asynchronously through email, annotations and so forth. Once again, the issue of scale needs to be considered - how can we support many tens, hundreds or even thousands of users in co-present interaction in a 3-D space? Recent Collaborative Virtual Environments (CVEs) have been used to hold wide area meetings up to ten simultaneous participants engaged in graphical and audio interaction (e.g. the recent European wide meeting held in our own MASSIVE system [Greenhalgh and Benford (1995)]). The next generation of CVEs aim to scale to a hundred or so simultaneous users through the use of spatial awareness techniques, multicast protocols and predictive behaviour modelling mechanisms. For example, we hope to achieve a significantly larger meeting in MASSIVE-2 by the Summer of 1996.
3. Implementation
We now describe the implementation of basic examples of these various techniques in order to construct a demonstration collaborative WWW interface. The following have been constructed as applications of, or extensions to, the DIVE (Distributed Virtual Environment) collaborative virtual reality system from the Swedish Institute of Computer Science (SICS) [Carlsson and Hagsand (1993)].
3.1 A browsing visualisation
Our browsing visualisation can provide an overview of up to a few hundred linked WWW nodes. file://I:\GMD-paper.html 3/8/00
Collaborative Visualization of Large Scale Hypermedia Databases (29-Mar-1996)
Page 5 of 7
The visualisation is created using the Force Directed Placement algorithm mentioned above and is called FDP-Grapher. Scoping in this visualisation involves seeding the visualisation with one or more starting objects; each representing the start of the search for objects to consider for inclusion in the visualization. The starting object(s) could be an organisation home page, a page of links relevant to a certain topic, the results of dynamically querying a search engine for a topic, or any combination of the above. Using a breadth-first algorithm, the search progresses outwards from the seed object until a certain link adjacency radius has been traversed. At this stage, the 3-D graph is constructed. Figure 2 shows a FDP visualization of a section of our WWW space, with two users present.
FIGURE 2. FDP-Grapher
3.2 A searching visualisation
Our searching visualisation is called VR-VIBE (see [Benfordet.al. (1995b)] for a full description). Unlike traditional text retrieval systems which only allow users to run a single keyword search at a time, VR-VIBE allows users to explore the effects of comparing and dynamically manipulating multiple simultaneous keyword searches. In essence, a number of keyword searches are defined, each of which consists of one or more text keywords. These are then positioned in a virtual space to form a spatial framework of queries. Document icons are positioned within this framework according to the strengths of their relative attractions to each query (i.e. the more strongly an individual document matches an individual query, the closer it is placed to it). In addition, the size and shade of a document icon shows its overall attraction to all of the queries. VR-VIBE users may dynamically interact with the visualisation in a number of ways: selecting documents displays summary details or launches Mosaic to view the document source if stored in the WWW (links are maintained from the local document repository into the Web); raising a relevance filter removes all documents whose overall score falls blow a threshold value from the display; grabbing and dropping queries dynamically re-arranges the space into a new configuration; switching queries on and off also changes the space and, finally, any number of new queries may be defined and positioned on the fly. Figure 3 shows VR-VIBE being used by several users to browse a document store, and find particular documents relevant to them.
FIGURE 3. VR-VIBE
3.3 Embodying and representing WWW users in the visualization
Given the use of the DIVE system, both of the above visualisations are shareable by directly embodied users who are able to communicate over external audio channels or via specialised meeting tools such as the DIVE mdraw whiteboard. In addition, we also provide 3-D representations of normal 2-D (e.g. Mosaic based) users. We have implemented a component called FollowWWW which links to HTTP servers or web browsers, and multicasts the position of users in the web to the FollowWWW server which collates the positions received from all sources, overlays a user embodiment on the visualization for each currently distinct browser, and moves the embodiments in real time to follow the movement of the user around the web. This process is represented in Figure 4.
file://I:\GMD-paper.html
3/8/00
Collaborative Visualization of Large Scale Hypermedia Databases (29-Mar-1996)
Page 6 of 7
FIGURE 4. Multicasting the positions of web users to the Visualization When connected to an HTTP server, the Fwww component allows the user browsing the space to register his or her name, email address, location of a facial image, etc. This information is subsequently used to enhance the representation of the user in the visualization. Figure 5 shows a group of users who have registered their personal details clustered around a web page and the pages linked from it.
4. A specific application: The Internet Foyer
We finish by describing one final aspect of our work - the configuration of the above collaborative visualisations as the Internet Foyer. The motivation for the Internet Foyer arises from two observations.
l foyers as physical entry points into real-world buildings serve several purposes. First, they
provide a formal entry point into an organisation. Second, they provide a location for generally useful information about the organisation. Third, they provide a meeting place. Fourth, they enhance security by making people pass through a publicly monitored space in order to gain access to other more private spaces (entrants may even have to formally sign in).
l an organisation’s home pages act as a kind of electronic foyer - an entry point into the
organisations electronic space. However, unlike with real foyers, people passing through are not directly visible to one another.
FIGURE 5. An overview of the Internet Foyer
FIGURE 6. A group of 2D users clustered around a web page
We argue that the kinds of visualisations described already constitute a kind of electronic foyer. However, we have recently taken this argument stage further. First, we have projected a browsing visualisation of our home pages onto the wall of a real ‘foyer’ using a video projection system attached to a Silicon Graphics workstation (actually, the foyer is currently only our laboratory). As a result people entering our WWW space (both 3-D DIVE users and 2-D Mosaic users) become publicly visible to people in the realfoyer much as anyone would who walked in off of the street. Second, we have texture mapped a video image of the physical foyer into a virtual window in the visualisation. Thus, DIVE users entering the visualisation can look through this virtual window and can see and interact with people in the real world. Thus, the whole system acts as a kind of hybrid space that we call the Internet Foyer; on one side of the wall is the real world; on the other is the WWW and its users, made visible as a 3 graphical environment! This represents an alternative -D way of thinking about collaborative access to the WWW where, in this case, we provide a publicly shareable interface which integrates WWW and realworld users into a single collaborative environment.
FIGURE 7. The Functionality of the Internet Foyer
5. References
file://I:\GMD-paper.html 3/8/00
Collaborative Visualization of Large Scale Hypermedia Databases (29-Mar-1996)
Page 7 of 7
Benford S, Bowers J, Fahlén LE, Greenhalgh C and Snowdon D (1995) User Embodiment in Collaborative Virtual Environments. In Proc. 1995 ACM Conference on Human Factors in Computing Systems, CHI'95, Denver, Colorado, USA 7-11 May 1995. ACM Press, pp. 242-249. Benford S, Snowdon D, Greenhalgh C, Ingram R, Knox I and Brown C (1995b) VR-VIBE: A Virtual Environment for Co-operative Information Retrieval. In Proc. Eurographics '95, Maastricht, The Netherlands, September 1995, pp. 349-360. Card S, Robertson GG and Mackinlay JD (1991) The Information Visualiser, an Information Workspace. In Proc. CHI'91, Human Factors in Computing Systems, ACM SICGHI, 1991, pp. 181-188. Carlsson C and Hagsand O (1993). DIVE - A Platform for Multi-User Virtual Environments. Computer & Graphics, Vol. 17 No. 6, pp. 663-669. Chalmers M and Chitson P (1992). Bead: Explorations in Information Visualisation. In Proc. SIGIR'92, published as a special issue of SIGIR forum, ACM Press June 1992, pp. 330 -337. Fruchterman TMJ and Reingold EM (1991). Graph Drawing by Force Directed Placement. Software Practice and Experience, Vol. 21 (11), pp. 1129-1164. Greenhalgh C and Benford S (1995). MASSIVE: A Collaborative Virtual Environment for Teleconferencing. In ACM Transactions of Computer Human Interaction, ACM Press Vol. 2(3), September 1995, pp. 230-261. Olsen KA, Korfhage RR, Sochats KM, Spring MB and Williams JG (1993) Visualisation of a Document Collection: The VIBE System. Information Processing and Management, Pergamon Press Ltd., Vol. 29 (1), pp. 69 -81.
file://I:\GMD-paper.html
3/8/00