Personalized Ontologies for Web Search and Caching

1
Personalized Ontologies for Web Search and Caching

• Susan Gauch
• Information and Telecommunications Technology
  Center
• Electrical Engineering and Computer Science
• The University of Kansas

2
Outline

• Motivation
• User profiles
• creation and maintenance
• evaluation
• Applications
• re-ranking (and filtering) search results
• Web caching
• Conclusions

3
Motivation

• Decrease access time for Web pages
• Server approaches
• use access logs to decrease access times for
  popular pages
• not tailored to individuals
• doesnt decrease network traffic
• Network approaches
• cache popular pages multiple places in the
  network
• not tailored to individuals

4
Personalization

• Different information needs for different users
• can we learn users interest?
• Explicitly?
• Implicitly
• can we use this information?
• improved search
• improved browsing
• faster Web page access

5
Intelligent Web Caching

• Improved (and faster) search results
• pre-caching all search results expensive
• Internet search engines return 50 irrelevant
  pages
• improved knowledge of users likely behavior
• intelligent pre-caching
• use past behaviors to predict future behaviors
• pre-cache best pages close to individuals

6
Context

• ProFusion www.profusion.com
• OBIWAN distributed content based IR
• Web clustered into regions
• clustering criteria content, location, company
• search query brokered to best regions within
  region brokered to most promising sites
• browsing a region means browsing its sites
  simultaneously
• www.ittc.ukases.edu/obiwan

7
User Profiles

• Applications
• Usenet news filtering
• recommendation services web browsing, books
• intelligent pre-caching
• Should
• accurately reflect actual interests
• require as little feedback as possible
• be dynamic

8
User profiles Creation

• Obvious and often used keywords
• not structured (ambiguous)
• static
• have to be explicitly mentioned
• Our approach
• watch over a user's shoulder while surfing
• automatically determine documents content
• central large ontology (concept hierarchy)

9
Document Classification

• Documents as weighted
• keyword vectors
• n different words-gt n dimensions
• weights based on
• word frequency and rarity
• Browsing hierarchy 10 web pages per node
• Concatenate them -gt keyword vector
• Content of a page most similar vector

10
Updating profiles

• Static document related
• content weights of top nodes for surfed document
• length of page
• Dynamic time spent
• Combine them
• for instanceweight (time/length)
• changes in interest in the five categories
• User profile weighted ontology

11
Profile evaluation

• Accordance with actual user interests
• 10/20 interest categories describe actual
  interests
• describe interests
• pretty well 3.5/5
• Convergence
• stabilization of ofcategories over time?
• do converge after 320
• surfed pages!

12
Profiles Summary

• Stored as weighted ontologies
• Profiles represent actual interests quite well
• Up to 150 top categories
• Two adjustment functions make profiles converge
• after 320 pages
• length of page doesn't really matter, but time
  spent does

13
Personalizing Search Results

• 50 of top 20 results irrelevant
• Same search mechanism for 200 million people?
• Goal
• identify relevant documents and put them on top
  of the result list
• (pre-fetch relevant results)
• Difficult problem 10 increase is very good

14
Re-Ranking

• Ranking a function of
• search engine's original ranking
• extents to which top 5 categories describe
  document's content
• personal interest in each of these top categories
• More relevant items on top of result list
• systems ability topresent all relevant items
• systems ability to present only relevant items

15
Recall and Precision

• Combination Recall/Precision graphs
• Example ranked documents 1,,20
• relevant 2,5,10,14,19
• recall points 1/5, 2/5, 3/5, 4/5, 5/5
• precisions 1/2, 2/5, 3/10, 4/14, 5/19

16
Re-Ranking Evaluation

• Overall performance increase of up to 8
• at each recall cutoff, up to 10 more relevant
  documents have been retrieved

17
Browsing Assistance

• Analyze current page
• locate links
• Identify which links are most likely to be
  followed by the user
• popularity of the link overall
• relevance of linked page to users interests
• Problem
• if you have to download the whole page to analyze
  it, youve increased the network utilization

18
Privacy

• Is the user aware that their behavior is being
  monitored?
• Can users turn it off?
• Where are profiles stored?
• With whom are profiles shared?
• How are profiles protected?
• How are profiles used?

19
Conclusions

• Automatic creation of structured user profiles is
  possible
• Profiles are reasonably accurate
• Applications in improving the search quality and
  Web page access efficiency
• Evaluation of re-ranking search results
  performance increase of up to 8

20
Future Work

• Incorporating profile generator into browser
• Connect system to ProFusion, OBIWAN
• Personalize structure of ontology
• Re-train classifier
• More applications recommendation service, web
  caching, browsing, ...
• Explicit user feedback?