A Novel Visualization Model for Web Search Results

1
A Novel Visualization Model for Web Search Results

• Nguyen T, and Zhang J. 2006. IEEE Transactions on
  Visualization and Computer Graphics
• PAWS Meeting
• Presented by Jae-wook Ahn
• February 9, 2007

2
Motivations

• Search result presentation
• Linear ranked list vs. visualization
• Interactivity
• Exploration
• Discover and analyze information by users
• Simple term based visualization models
• Multiple meanings
• Order of keywords
• Missing context information

3
Motivations (contd)

• What is missing with current approaches?
• Semantic views/semantic relations
• Degree of relevance in terms of subjects of
  interest
• Visualization model adapting to users subjects
  of interest and contextual information

4
Proposed approach

• Metaphor solar system
• Query Sun
• Documents planets
• Dimensions and attributes
• Semantic strength ( relevance)
• Gravity (distance)
• Rotation speed
• Color
• What is unique?
• Use movement, speed, and distance to visualize
  the degree of relevance among a query and Web
  search results with respect to users subjects of
  interest and contextual information

5
Architecture

• WebSearchViz Java based meta search (Google)
  visualization
• Vector space model document-term matrix
• Term weighting
• Similarity measure

6
Subject of interest

• Subject List of keywords
• Users can manually add/remove/edit subject of
  interests
• Weights are adjustable

7
Visualization space
Subject of interest

• Location of documents
• Similarity to the query
• More similar, closer
• Similarity to the subjects
• More similar, closer
• Rotation speed of the documents
• Similarity to the subjects
• More similar, more identical to the speed of the
  subjects
• Computed by the angles (bi)
• Rotation
• Automatic or manual

Document
Query
8
WebSearchViz
(1) Googles search results
(3) Visualization
(2) Users subjects
(4) Rotation control
9
Additional features

• Manual rotation
• Users can view the impact of a moving subject
• Colors
• Users can mark a document and can keep track it
  through sessions
• Center switching
• Any page can be made the center
• Filtering
• Users select a threshold by creating a filter
  circle
• Filter out low similarity documents
• Grouping
• Visualize only a group of documents

10
Experiments

• Time efficiency
• 0.7 sec (parsing indexing)
• 0.5 sec (document-term matrix, 10,000 page,
  54,000 keywords)
• Initial Visualization rendering 1.2 sec
• Usability study
• 20 undergraduates (natural social science)
• Questionnaire accuracy evaluation
• 89 recall, 92 precision
• More successful with WebSearchViz
• Subjects liked more manual rotation
• Subjects liked grouping, filtering, focus page
  shifting
• Short learning curve enhancement (WebSearchViz)
  to an existing search service (Google)

11
Conclusions

• Web search visualization
• Considers context subjects of interests
• Subject editor
• New dimension rotation speed
• Not just decorative