Query Expansion

1
Query Expansion

• By Sean McGettrick

2
What is Query Expansion?

• Query Expansion is the term given when a search
  engine adding search terms to a users weighted
  search.
• The goal is to improve precision and/or recall.
• Example User Query car Expanded Query car
  cars automobile automobiles auto etc

3
Classes of Query Expansion

• Human and/or computer generated thesauri
• Relevance feedback
• Automatic query expansion

4
Query Expansion Issues

• Two major issues
• Which terms to include?
• Which terms to weight more?
• Concept-Based vs. Term-Based Query Expansion
• Is it better to expand based upon the individual
  terms in the query, or the overall concept of the
  query?

5
Relevance of Query Expansion

• Query expansion is very important on the web.
• The amount of information on the web is always
  increasing.
• In 1999, Google had 135 million pages. It now
  has over 3 billion.
• Search engine users follow specific trends with
  their searches.
• 2-3 words
• Broad search term
• Do not like to expand their queries either
  through refining search terms or using Boolean
  operators

6
Thesauri

• What is a Thesauri in the IR world?
• Any data structure that defines semantic
  relatedness between words.
• Schutze and Pedersen (1997)
• Often more complex than normal Thesauri.
• Thought to be too broad to be useful.

7
The Need For Thesauri

• Naturally assumed that pulling words from a
  thesauri would increase
• The number of documents retrieved.
• Possibly precision.
• The car example car vs. car, auto,
  automobile, vehicle, sedan, etc
• Which would retrieve the largest number of
  documents?
• Is larger necessarily better?

8
Human Automatically Generated Thesauri

• Earliest work began in the 1950s.
• H.P. Luhn
• Thesaurofacet detailed list of engineering
  terms
• Largely used in such industries as medicine,
  aerospace, and other technological fields.

9
Drawbacks of Handcrafted Thesauri

• Cost
• Development.
• Maintenance.
• Cost often outweighs benefit.
• Time
• It often takes a long time for thesauri to
  develop.
• Hard to keep up with the pace of scientific and
  technological development.

10
Automatically Generated Thesauri

• Need grew from limitations of handcrafted
  thesauri.
• No longer the cost of experts to generate
  thesauri.

11
Automatically Generated Thesauri

• 3 Steps.
• Extract word co-occurrences.
• Define word similarities.
• Based upon word co-occurrence or lexical
  relationship.
• Cluster words based upon their similarities.
• Not proven very successful.
• As late as 1990 many industries were still using
  handcrafted thesauri.

12
Relevance Feedback

• Began in the 1960s.
• Significant improvement in recall and precision
  over early query expansion work.
• Basic process as follows.
• The user creates their initial query which
  returns an initial result set.
• The user then selects a list of documents that
  are relevant to their search.
• The system then re-weights and/or expands the
  query based upon the terms in the documents.

13
Relevance Feedback Models

• Many different types of models.
• Depend on methods and theories behind them.
• Vector Space.
• Probabilistic.
• Boolean.

14
Ide dec-hi Method

• In this method, all the top ranked relevant
  documents are used as is the highest ranked
  non-relevant document.
• The non-relevant document is used a point in the
  vector space from which the feedback query is
  removed.
• Up to 160 improvement over non-expanded queries.

15
Interactive Query Expansion

• Uses a thesaurus.
• After initial query is submitted, the system
  returns a list of associated and relevant words
  derived from both the result set and a thesaurus.
• Useful, but more research is needed.

16
Pseudo-relevance Feedback

• Grew from problems involved in implementing
  relevance feedback systems.
• Users do not like to give manual feedback to the
  system.

17
Pseudo-relevance Feedback Process

• The system returns an initial set of documents.
• The system assumes that the top n number of
  documents are relevant to the query.
• The system takes terms from these documents to
  re-weight the query.
• Relies largely on the systems ability to
  initially retrieve relevant documents.

18
lol
19
Automatic Query Expansion

• The process of automatic query expansion using
  computer generated thesauri.
• Works somewhat like pseudo-relevance feedback.
• Implementation not as useful, but still widely
  researched.

20
Term Co-occurrence Measures

• Process of developing relationships between
  words based upon their co-occurrence in
  documents.
• Clustering
• Documents that share a significant number of
  terms are grouped together.
• A thesaurus is then generated from the terms in
  these categories.
• Categories sometimes too narrow or broad.
• Does not account for synonyms.

21
Lexical Co-Occurrence Measures

• Instead of looking at the frequency of terms in
  a document, the proximity of words in a document
  is looked at.
• Context of words becomes important.
• Some performance improvement shown in small
  document collections.
• Not quite as good as relevance feedback, but
  better than pseudo-relevance feedback.

22
Current State of Query Expansion

• Query Expansion technology has reached somewhat
  of a plateau.
• This is due to limiting factors of relevance
  feedback and word co-occurrence.
• Current research attempting to refine previous
  research in the field.

23
Where To Go From Here?

• Grammatical Based Thesauri
• Syntactical relationship between words
• Words placed into classes
• Some improvement on small document collections.
  Failed on larger ones.
• AI Searching
• Mostly theory
• Intelligent Agents
• Could be customized reflect specific needs of the
  user
• Next logical step in IR, but still far off from
  commercial use

24
Works Cited

• Attardi, G., S. Di Marco and F. Sebastiani. 1998.
  Automated Generation of Category-Specific
  Thesauri for Interactive Query Expansion.
• Grefenstette, G. 1992. Use of Syntactic Context
  to Produce Term Association Lists for Text
  Retrieval. In Proceedings of the 15th Annual
  International ACM-SIGIR Conference on Research
  and Development in Information Retrieval,
  Copenhagen, Denmark, ed. N. Belkin, P. Ingwersen
  and A. M. Pesjtersen pp. 89-97. New York ACM
  Press.
• Ide, E. 1971. New Experiments in Relevance
  Feedback. In G. Salton. The SMART Retrieval
  System Experiments in automatic document
  processing. Englewood Cliffs, NJ Prentice-Hall.
• Qiu, Y., 1993. Concept Based Query Expansion. In
  Proceedings of SIGIR-93, 16th ACM International
  Conference on Research and Development in
  Information Retrieval.
• Schutze, H. and J. Pederson. 1997. A
  Cooccurance-based Thesaurus and Two Applications
  to Information Retrieval. Information Processing
  and Management 33, no. 3 pp. 307-318.
• Walker, D. 2001. Query Expansion Using Thesauri.