Navigationbased Information Seeking with User Personalization

1
Navigation-based Information Seeking with User
Personalization

• Yuanhua Lv
• Nov. 28, 2007

2
Outline

• Problems with V.3
• Navigation-based Information Seeking
• User Personalization

3
Problems with V.3

• Inefficient and inconvenient information seeking
• query -gt map -gt map -gtmap
• query1.0, query2.0, query3.0,
• save space1, save space2, then manipulate them
• switching between subspaces?
• complicated user interface
• One-fit-all
• Lacking user modeling, not adaptive to individual
  users

4
Navigation-based Information Seeking
Querying
Topic Map for Browsing
Topic Region
Navigating from a node to a child node ? Map
Navigating from a node to its neighbor ? Space
Switching
5
Navigation-based Information Seeking

• Integrating Browsing and Querying in a
  Navigation-based information seeking framework,
  treating both as navigation over topic regions.
• Supporting more collaboration between Browsing
  and Querying
• Help users do space switching and space algebra
  easily

6
Building a multi-resolution topic map

• Possible Solutions
• Forming topic regions user-defined, ontology,
  clustering?
• Generating topic region relations
• Vertical relations (children and parents)
  hierarchical clustering?
• Horizontal relations (neighbors) frequent access
  pattern (People that searched behavior1 also
  searched behavior2), content similarity?
• Labeling map node an updated query for
  particular Space? Qiaozhus work?

7
Ranking

• Possible Solutions
• Ranking the search results of a topic region
• Context-sensitive utilizing rich context
  information, including queries, clickthroughs,
  browsing actions such as zoom in/out operations
  and neighborhood explorations
• Ranking the topic regions of a map
• Eager Feedback based on the users immediate
  implicit feedback information (Xuehua)

More sophisticated strategy User Personalization
8
User Personalization

• Most existing information retrieval systems,
  including the web search engines (e.g. Google,
  Yahoo!), share the common problem of one size
  fits all they do a user-generic retrieval that
  tries to serve everyone, and therefore never
  satisfies anyone. J. Allan et. al, 2003

9
Jaguar
Adapted from X. Shens SIGIR05 presentation
10
User Personalization

• Recently there have been many attempts to address
  this issue through exploiting users personalized
  information.
• K. Sugiyama et al. (2004) and J. Teevan et al.
  (2005) used all available user information to
  personalize search results.
• A. Pretschner S. Gauch (1999) and F. Liu et al.
  (2002) grouped user information into categories,
  and tried to restrict search results to these
  categories.
• B. Tan et al. (2006) exploited relevant query
  records that are similar to the current search
  results to update query model

Noisy Data
11
User Personalization

• (Ongoing) We try to reduce noisy data through
  exploiting long-term query chains, which
  satisfies the following constraints
• Semantically Close relevant to the current query
• Context Sensitive consistent with the context
  information
• Discriminative across the current search results
  high relevance to one topic and low relevance to
  other topics
• The long-term query chains would be applied to
  the ranking problem of the navigation-based
  information seeking framework.

12
Summary

• Navigation-based Information Seeking
• User Personalization

13
Discussion

• What exactly should the navigation interface look
  like?
• What are some basic operations? (e.g., querying,
  browsing, what else?)
• How do we support a user to manage a personalized
  information space (e.g., MyTopics, MyGenes, etc)
• What user actions should we log?