Ed H. Chi www.geekbiker.com

1
Ed H. Chi www.geekbiker.com

• U of Minnesota
• Ph.D. Visualization Spreadsheets
• M.S. Computational Biology
• Expertise InfoVis, Study of the Web, TaeKwonDo,
  Poetry, Motorcycling, Pottery

2
Information Scent Modeling User Browsing
Strategies on the Web

• Ed H. Chi
• Peter Pirolli
• User Interface Research Group
• This research was supported in part by
• Office of Naval Research contract number
  'N00014-96-C-007'.

3
Comparison to Library

• Experience tells us
• general layout of content
• which floor, which section.
• which books are of greatest interest
• by the wear on the spines.
• which information is timely or deadwood
• by looking at the circulation check-out stamps
  inside the book covers.

4
Trends and Problems

• 200M Web users, 6M web sites
• Web design process ad-hoc, not optimal
• Some tools extract behaviors and correlations but
  not intentionally
• Being successful requires making the Web more
  useful and usable to a broader audience

5
Information Foraging
6
Underlying Concept

• Users seeking information is similar to
  hunter/gatherers optimization strategies.

7
Underlying Concept

• Information Scent is the user perception of the
  cost and value of information.
• Similar to hunters following animal foot prints.

8
(No Transcript)
9
Information Scent

• Users forage by surfing along links
• Foragers use
• proximal cues (text snippets or graphics) to
  access
• distal content (destination page)
• Scent is the proximal perception of value and
  cost of distal content

snippet
content
link
10
Assumptions

• Users have information goals, their surfing
  patterns are guided by information scent
• Two questions
• Given an information goal and a starting point
• Where do users go? (Behavior)
• Given some surfing pattern
• What is the users goal? (Need)

11
WUFIS Web User Flow by Information Scent
Web site
Web Page content
links
User Information goal
Web user flow simulation
Predicted paths
12
How does it work?
Start users at page with some goal
Examine user patterns
Flow users through the network
13
WUFIS Algorithm
1
Relevant Documents
Weight Matrix
Query
14
WUFIS Algorithm (cont.)
2
Scent Matrix
R Relevant documents T Topology matrix
15
Prelim. Evaluation of WUFIS

• Show that WUFIS generates good URL destinations
  based on information need.
• 19 Websites
• Size 27-12,000 pages
• Info Provider, eCommerce, Large Corp.
• Info Need from very general (product info) to
  very specific (migraine headaches)
• Top ten URL position simulated are extracted.
• Each URL is blindly rated for relevancy.

16
WUFIS Evaluation

• 570 ratings are collected 3 variations of the
  algorithm x 10 URLs x 19 sites
• Tabulated, Averaged.
• Result 7.54 (out of 10)

19 Websites
Website Info, Algorithm Performance
17
IUNIS Inferring User Need by Info Scent
Web site
Web Page content
links
User Information goal
Web user flow simulation
observed paths
18
Extracting Paths

• Longest Repeating Sequence (LRS)
• New path mining technique
• Extracts significant surfing paths
• Reduces the complexity of path model

19
IUNIS
1
P observed user path T topology matrix W
word x document weights K relevant keywords
2
Topology
Path
Weight
Path
20
Evaluation of IUNIS

• Goal
• Show that keyword summaries produced by IUNIS are
  good at communicating the content of the user
  paths.
• Dataset
• 8 participants
• random 10 paths from (5/18/1998, xerox.com, path
  length6)
• booklets of pages on paths (in order)

21
Evaluation of IUNIS

• Procedure
• Single rating sheet with the ten 20-word
  summaries. Beside each summary, users are asked
  to rate the summaries on a 5-point Likert Scale.
  A copy of this rating sheet is attached to each
  of the ten path booklets
• Users are asked to read through each booklet and
  rate each of the path summaries.
• User are also asked to identify which of the ten
  summaries was the best match.

22
Evaluation of IUNIS

• Results
• Matching summary mean 4.58 (median5)
• Non-matching summary mean 1.97 (median1)
• Difference highly significant (p lt .001)
• Best match summary 5.6 out of 10 (Cohen
  Kappa0.51)
• Evaluation yield strong evidence that IUNIS
  generates good summaries of the Web paths.

23
ScentViz Tasks

• Overall site
• High-level traffic flow and routes?
• Ease of access and costs?
• Given a specific Web page
• Where do users come from?
• Where do they go?
• What other pages are related?
• Users
• What are interests of the users?
• Where should they go based on their need?
• Do observed data match simulation?

24
Visualization Demo

• Dome Tree
• Usage Based Layout
• Path Embedding

25
Scenario 1 Page Types

• Multi-way branching point

investor/sitemap.htm
26
Scenario 1 Drill-down

• Few well-traveled future paths
• shareholder info
• 1998 fact book
• financial doc order
• Conclusion
• good local sitemap

27
Scenario 2 Well-traveled

• Related information all over the site
• One well-worn path on the left relating to
  product tutorial

Scansoft/tbpro98win/index.htm
28
Scenario 3 Identify Need

• Need of path from shareinfo to orderdoc
• reinvestment
• stock
• brochure
• dividend
• shareholder

investor/sitemap.htm
29
Scenario 4 Scent Predict

• Scent computed based on pagis need
• Good match between scent and LRS paths

Scansoft/pagis/index.html
30
InfoScent Summary

• The overall goal is to model Web user information
  needs
• Bridge gap between clicks and information needs
• Predict user navigation behavior
• Develop new applications and Web usability metrics

31
Questions?

• Ed H. Chi
• Chi_at_acm.org
• http//www.geekbiker.com