Design Lessons from Deployment of On-demand Video
Liwei He, Anoop Gupta, Stephen A. White, and Jonathan Grudin
Microsoft Research
Redmond, WA 98052
{lhe,anoop,stevewh,jgrudin}@microsoft.com
ABSTRACT detailed logs of system use.
Streaming video to the desktop is increasingly widespread. The interface to a technical talk is three web-based frames:
A key application is in training, making information video of the presenter in the top left, a table of contents
available over the Internet or corporate intranets, in real (TOC) in the bottom left, and the speaker‟s PowerPoint
time or as archived presentations. How should a slides in a large frame on the right.
presentation be redesigned for retrieval and viewing on
demand? Detailed examination of usage logs of 6000 The slides change automatically in synch with the video.
corporate on-demand video sessions provides suggestions. The TOC contains one line for each slide title. When a
TOC entry is clicked on, the presentation jumps to the
Keywords corresponding spot in the talk. Audience questions are
Video on-demand, streaming media, digital library added to the TOC during the post-presentation preparation
INTRODUCTION of the archived digitized video.
With the steady rise in desktop computer performance and MSTE USE
network bandwidth, on-demand digitized video will be From July 1997 to June 1998, 163 talks were digitized.
routinely available in workplaces and educational settings. Most were between 30 minutes and 2 hours. Over 2000
Digitized video presentations are not without cost and do employees accessed archived presentations, 6685 viewing
not provide the same opportunities for direct interaction
sessions in all.
with a speaker and audience, but they provides several
compensatory advantages. The content can be watched Every viewer interaction with an interface element, such as
anywhere and anytime by audiences of any size. VCR-like TOC entries or play, stop, and other video controls, is
operations and advanced compression and indexing logged. These records, numbering over 115,000, are the
methods can be made available. Viewers can time-share basis of our analyses, together with some interviews.
activities, selectively watch sections, or „leave‟ without 1400
offending a speaker. One-time users
1200 2-to-4-time users
A single presentation may now be given to a live audience,
Number of sessions
5-to-9-time users
with other viewers watching from a distance through 1000
10-plus-time users
videoconferencing or desktop videoconferencing, and yet 800
other viewers watching later via archived video on demand.
The presenter may have to develop materials and 600
presentation style with all three audiences in mind. 400
In this paper, we focus on on-demand viewing. Just as 200
designing text for effective viewing on the Web differs
from text designed for other reading situations [1,2], 0
speakers may wish to tailor presentations specially for on-
Ju 8
D 97
Ap 8
Fe 98
98
Au 97
Ja 7
M 8
Se 97
O 7
M 98
N 97
-9
-9
-9
9
9
-
n-
n-
l-
b-
g-
p-
r-
-
demand viewing.
ay
ar
ov
ec
ct
Ju
THE SYSTEM
Figure 1: User sessions by month and user type.
Microsoft Technical Education (MSTE) provides internal
technical education on a range of topics to employees. Over Figure 1 shows the number of sessions by month (the drop
a year ago MSTE started making digitized videos of these in November and December reflects an interruption in
talks available on an intranet web site and collecting logging during a system change and break in operation).
Users are segmented into those who only accessed one
video all year, those who accessed 2-4, 5-9, and 10 or more.
These data indicate that MSTE steadily drew first-time
users. In fact, 1234 of 2096 users only used MSTE once.
However, the other 800 viewers accounted for over 80% of
the accesses, so the results below are not inordinately due to
one-time use.
4
Average number of sessions per talk
presenters must pack an overview and major points into the
3.5 first several minutes.
3 0.60
2.5 One-time users
0.50
Fraction of total sessions
2 2-to-4-time users
5-to-9-time users
1.5 0.40
10-plus-time users
1
0.30
0.5
0.20
0
1
4
7
10
13
16
19
22
25
28
31
34
37
40
43
46
49
52
Number of weeks after live presentation 0.10
Figure 2: Average number of distinct users accessing a talk as 0.00
a function of weeks since the presentation was given. 0 1-4 5 - 25 25 or more
Is the viewing of a talk concentrated in the days following Number of jumps
its presentation? Figure 2 shows that this is not strongly the Figure 5: Jump (TOC use) statistics.
case. This has implications for operation (archiving video People do use the TOC to index or jump into the video, and
takes considerable storage) and also for presentation design. their use does not vary significantly for light and heavy
0.6 MSTE users (Figure 5).
One-time users This also has design implications. In a live presentation, a
Fraction of total sessions
0.5
2-to-4-time users
5-to-9-time users
slide can be understood by its context: its title can be
0.4
10-plus-time users whimsical or elaborated by the speaker. MSTE users rely
0.3 more heavily on slide titles and slide contents when they
inspect the TOC and jump from slide to slide. Presenters
0.2
should put much more thought into slide titles and the
0.1 clarity of slide organization and emphasis. They might also
0
put more content on slides, sacrificing a little clutter for the
0-5 5 - 25 25 - 50 50 - 75 75 - 100 100+ live audience to provide clarity to later viewers.
Normalized session length (session length / talk length)
CONCLUSIONS AND FUTURE WORK
Figure 3: Session duration by length of talks. MSTE use provides evidence that video on demand is used
Another interesting question is what percentage of a talk and may be worth the effort to make it available. We also
was watched? As shown in Figure 3, 51% of the sessions find strong evidence that effective design of video
are less than 5% of the talk length. Although many longer presentations for access on the Web, like designing text for
sessions occurred, this surprised us. In addition, there is not the Web, requires new approaches based on understanding
much difference depending on frequency of accessing the use of this potentially powerful educational medium.
MSTE, although the most frequent (10+) viewers may be A longer version of this report may be accessed at
more likely to watch most or all of a video. http://www.research.microsoft.com/pubs/ (search for “on-
0.7 demand video”). We are presently extending these analyses,
interviewing MSTE users, and extending the duration of the
Fraction of sessions that cover the
0.6
study.
0.5
ACKNOWLEDGMENTS
0.4 Thanks to the Microsoft Technical Education Group for
minute
0.3
giving us the data and Mary Czerwinski who reviewed the
paper and gave us valuable suggestions for improvement.
0.2
0.1
REFERENCES
0
1. Morkes, J., and Nielsen, J. “Concise, SCANNABLE,
1
10
19
28
37
46
55
64
73
82
91
100
109
118
127
136
145
154
and Objective: How to Write for the Web.”
Nth minute
http://www.useit.com/papers/webwriting/writing.html.
Figure 4: Viewing likelihood minute by minute.
2. Nielsen, J. “Guidelines for Multimedia on the Web.”
Figure 4 shows the likelihood that a given minute of the
Sun, 1997. http://www/w3j.com/5/s3/nielsen.html.
video is being watched. Figures 3 and 4 have design
implications. A live audience is less likely to walk out five
minutes into a presentation. For on-demand viewing,