IST 497G Ron Grzywacz November 2002

1
IST 497GRon GrzywaczNovember 2002

• Personalization of Information Retrieval

2
(No Transcript)
3
(No Transcript)
4
Overview

• The Topic
• Issues
• Importance
• What has been done and how
• The next step

5
Personalization of IR

• This refers to the automatic adjustment of
  information content, structure, and presentation
  tailored to an individual user.
• Characteristics
• Age
• Gender
• Special Interest Groups
• Topic

6
Issues

• Why Use?
• One Size Doesnt fit all
• Limits Diversity
• Limits Functionality
• Limits Competition
• Different Users have different needs

7
Issues

• How to customize the process?
• Which characteristics?
• Not all people of certain groups are the same
• Virtually impossible to create one unique search
  engine for every individual
• Wouldnt make sense to build 100 million
  different versions of Google from the group up
  based on one users needs

8
Importance

• Why important?
• Not every solution is ideal for each person
• Certain people dont understand how to use
  certain systems
• Current systems arent tailored for a specific
  type of user

9
The Example

• Suppose we query a system for Michael Jordan
• Most popular engines would return information
  about the famous basketball player
• Suppose that we were looking for information
  about computer science papers written by Michael
  Jordan
• By using the query Michael Jordan, there is no
  information regarding the context of our desired
  search

10
Context Search

• This presents a problem
• We have no way to infer or assume the context of
  a users desired search
• Currently its a hit or miss process to return
  relevant contextual information
• What if we tried to automatically infer
  contextual information?

11
Automatic Inferring of Contextual Information

• By monitoring user patterns, we could gather
  information about the user and the possible
  context of search
• This raises issues regarding privacy
• Do you want some company building information
  about your preferences?
• What happened if this information was released or
  misused against you.

12
Personalized Search

• By combining the previous items we can build a
  personalized search service.
• The example Michael Jordan query could return
  data about the basketball player to sports
  enthusiasts and information about computer
  science papers to researchers.

13
Whats Been Done?

• Inquirus/Inquirus 2
• This is a meta-search engine done by a team from
  NEC (which included Prof. Giles)
• It attempts to add a category or contextual
  information to a keyword search
• Examples would be personal homepage, research
  paper and general introductory information
• It uses the information to query relevant search
  engines, modify queries and select ordering policy

14
Whats Been Done?

• My experiences with Inquirus
• You have 3 search options within this service
• Web Search
• Google
• Groups (Google Groups)
• Returned valid results from both Google and
  Groups
• Valid results were also returned from Web Search
  but rank order was not as good
• It also included words such as of in the search
  terms. This could be problematic, since sometimes
  words like that had the most results

15
Whats Been Done?

• The Watson Project Northwestern Univ.
• Suppose someone is searching for information
  about cats
• Its easy to manipulate text to create many
  unique scenarios
• Veterinary student writing a term paper on animal
  cancer
• Feline Cancer, Diagnosis, Treatment
• Contractor working on a proposal for a new
  building.
• Caterpillar Corp. machinery
• Grade-school student writing a paper about Egypt.
• Pictures and stories about cat mummies and gods

16
Whats Been Done?

• The Problems with the query
• Relevance of active goals
• The active goals of the user contribute
  significantly to the interpretation of the query
  and to the criteria for judging a resource
  relevant to the query.
• Word-sense ambiguity
• The word sense of cat is different from the
  others in each scenario. The context of the
  request provides a clear choice of word sense.
• Audience appropriateness
• The audiences in each of the scenarios also
  constrain the choice of results. Sources
  appropriate for a veterinarian probably will not
  be appropriate for a student in grade school.

17
Whats Been Done?

• The Watson Project Solution
• A system used to collect contextual information
  from everyday computer use
• Watson is a client side application that monitors
  you daily computer use of applications such as
  word processors, web browsers and Email clients.
• By knowing about you and your work, Watson can
  help you find information that is relevant to you.

18
Whats Been Done?

• My experience with Watson
• A small download and brief install loaded the
  java based application
• I used it while creating this presentation
• It managed to return results regarding search
  engines and their development, but it did not
  return anything relevant to my specific topic
• It did manager to generate a search in CiteSeer
  for me

19
Whats Been Done

• Context already assumed
• Although we can not automatically assume the
  context of a users search yet, people have built
  engines that use a given context
• CiteSeer
• This is a search engine for research papers in
  scientific literature

20
Whats Been Done

• PubMed
• Customized Science and Medicine database of
  journals and articles
• Questia
• Search engine for students who are doing research
  and writing papers

21
Whats Been Done

• A different approach using personalized IR
• In what other ways can we use this type of
  technology?
• KnowledgeFlow Inc. Web Angel
• Browser Plug-in that stores your preferences and
  returns customized advertisements to you as you
  browse the web
• A practical consumer/commercial application of
  personalized IR

22
Whats Been Done

• There are other ways to personalize web content
• Recommender Systems
• A System that monitors your habits or receives
  input about your preferences and generates things
  you might be interested in
• Amazon.com
• Barnes and Noble

23
Other Recommender Systems

• Movielens
• Movie selection service
• Gives you a survey to gauge your preferences
• Makes recommendations based on your likes
• Book Forager
• Novel selection service
• Allows you to choose a variety of book
  characteristics
• Makes a recommendation based on your current
  choices

24
Whats Next?

• How can we build upon current services?
• Use of AI to evaluate query to determine
  contextual information
• Use user provided information to generate
  specific data
• Build upon user provided data by monitoring
  browsing preferences

25
Summary

• The Topic
• Issues
• Importance
• What has been done and how
• The next step

26
Citations

• Budzik, J., and Hammond, K. User Interactions
  with Everyday Applications as Context for
  Just-in-time Information Access. In Proceedings
  of Intelligent User Interfaces 2000. ACM Press,
  2000. (Nominated for Best Paper Award)
• Steve Lawrence. Context in Web Search, IEEE Data
  Engineering Bulletin, Volume 23, Number 3, pp.
  2532, 2000.
• N. Ramakrishnan and S. Perugini. The Partial
  Evaluation Approach to Information
  Personalization. ACM Transactions on Information
  Systems, August 2001.