Query Expansion and Context: Thoughts on Language, Meaning and Knowledge Organisation
4 likes697 views
This document summarizes a workshop on digital information management that took place in April 2012 in Corfu, Greece. It discusses some of the challenges of information retrieval when using natural language queries, including problems of ambiguity, context, and the use of knowledge organization systems and query expansion to help address these challenges. The role of user models and evaluation in understanding real language use is also mentioned.
![ […]for it’s not an underlying logic that clarifies what we
mean, it’s the context, activities and practices in which we
use language that provide the fundamental clarification of
meaning we are looking for (Blair, 2006)
Only in the stream of thought and life do words have
meaning. (L. Wittgenstein, “Zettel”, §173)
Meaning depends on consistent usage but requires more than
that; it also requires that speakers be able to check that
someone’s usage is consistent (Jaworski, 2011)
6](https://image.slidesharecdn.com/mastoradim2012-120510010723-phpapp02/85/Query-Expansion-and-Context-Thoughts-on-Language-Meaning-and-Knowledge-Organisation-6-320.jpg)
![ Since language is involved in their creation and
development, KOSs bear themselves the inherent
characteristics of the use of language
◦ Ambiguity [PoS: “looks”, Semantic: “bank”, Syntactic: “He hit
the girl with the hat”]
◦ Homonymy [Homophones: “too /two”, Homographs: “tire”]
◦ Polysemy [“mouth” (on the face OR the opening of a cave)]
◦ Synonymy [“sick – ill”]
19](https://image.slidesharecdn.com/mastoradim2012-120510010723-phpapp02/85/Query-Expansion-and-Context-Thoughts-on-Language-Meaning-and-Knowledge-Organisation-19-320.jpg)
![ Even if KOSs try to capture and deliver the absolute
meaning [or a more targeted one] of what they
describe, they still are considered “collection
independent knowledge structures” (Efthimiadis, 1996)
◦ There still are missing parts for the communication
of the intended meaning
Ambiguity differs only by degree between universal
and domain-specific classifications, though that
difference of degree is likely quite significant
(Szostak, 2010)
20](https://image.slidesharecdn.com/mastoradim2012-120510010723-phpapp02/85/Query-Expansion-and-Context-Thoughts-on-Language-Meaning-and-Knowledge-Organisation-20-320.jpg)
![Thank you!
Contact:
Anna Mastora
mastora [at] ionio [dot] gr
This research has been co-financed by the European Union (European Social Fund
– ESF) and Greek national funds through the Operational Program "Education and
Lifelong Learning" of the National Strategic Reference Framework (NSRF) -
Research Funding Program: Heracleitus II. Investing in knowledge society through
the European Social Fund.
22](https://image.slidesharecdn.com/mastoradim2012-120510010723-phpapp02/85/Query-Expansion-and-Context-Thoughts-on-Language-Meaning-and-Knowledge-Organisation-22-320.jpg)
Query Expansion and Context: Thoughts on Language, Meaning and Knowledge Organisation
- 1. Anna Mastora & Sarantos Kapidakis Laboratory on Digital Libraries and Electronic Publishing Department of Archives and Library Science Ionian University 2nd Workshop on Digital Information Management 25-26 April 2012, Corfu, Greece
- 2. Introduction Research hypothesis Language & Problems in Information Retrieval Query Expansion Context Knowledge Organisation Systems (KOS) Considerations 2
- 3. Information Retrieval (IR) is about retrieving relevant results as response to expressed information needs The expression of information needs uses natural language representations ◦ Language is clearly ambiguous (Szostak, 2010) Not all linguistic representations are distinguishably informative of the user’s intent 3
- 4. Description of terms ◦ Deals with each term individually e.g. definitions in a dictionary Discrimination of terms ◦ Deals with the relations between terms e.g. hierarchical relationships within a thesaurus (Blair, 2006) 4
- 5. By studying the actual use of language in every-day activities is how we can clarify meaning; deriving information about the context within which words are used will help us clarify better the identified indeterminacies 5
- 6. […]for it’s not an underlying logic that clarifies what we mean, it’s the context, activities and practices in which we use language that provide the fundamental clarification of meaning we are looking for (Blair, 2006) Only in the stream of thought and life do words have meaning. (L. Wittgenstein, “Zettel”, §173) Meaning depends on consistent usage but requires more than that; it also requires that speakers be able to check that someone’s usage is consistent (Jaworski, 2011) 6
- 7. 7
- 8. Language is a labyrinth of paths. You approach from one side and know your way about; you approach the same place from another side and no longer know your way about. (§203) L. Wittgenstein “Philosophical Investigations” (1967) 8
- 9. In order for Information Systems to deal with the problems caused by the diversity of linguistic representations, Query Expansion is implemented. ◦ Supplementing the original query with additional - meaningful- words or phrases (manually, automatically, semi-automatically) What *meaningful* means? More information about the “context” What is the *context*? 9
- 10. More information about the involved parties within the information retrieval process The Multi-faceted concept of Context (Bhatia & Kumar, 2010)
- 11. 11
- 12. 12
- 13. 13
- 14. - asking the users explicitly - statistical processing of log files (e.g. Query chains) - qualitative evaluation of log files, data or systems - pre-processing of document corpora - machine learning techniques (un-, semi-, supervised) - implementation of user behaviour models - personalisation/ profiling of users - knowledge organisation structures - ... 14
- 15. Meant to be some kind of a problem solver; a mediator between the inquirer and the content which is stored in the documents ◦ Placing a concept within a hierarchical definition establishes what sort of thing this is and what sort of thing it is not, and often sorts the subsidiary elements of which it may be comprised (Szostak, 2010) 15
- 16. The expert does not hold a “more complete” definition of the words than we do. He simply knows more about certain words than we do, and by providing this additional information about them, it may be useful for identifying each word in different circumstances. (Blair, 2006) 16
- 17. Schemes for organising information & promoting knowledge management (Hodge, 2000) ◦ Term lists (authority files, glossaries, dictionaries, gazetteers) ◦ Classification & categories (classification scheme, taxonomy, subject headings) ◦ Relationship lists (thesaurus, semantic network, ontology) Constitute ways for formalising knowledge and, consequently, communication through linguistic expressions or any other kind of definitional or descriptive sign, like visual. Used in Query Expansion to derive context 17
- 18. The epistemological basis of any theory of Knowledge Organization is an accepted postulate. In other words, how knowledge is organized and represented depends largely on the understanding of how knowledge is organized and represented. (Alexiev & Marksbury, 2010) 18
- 19. Since language is involved in their creation and development, KOSs bear themselves the inherent characteristics of the use of language ◦ Ambiguity [PoS: “looks”, Semantic: “bank”, Syntactic: “He hit the girl with the hat”] ◦ Homonymy [Homophones: “too /two”, Homographs: “tire”] ◦ Polysemy [“mouth” (on the face OR the opening of a cave)] ◦ Synonymy [“sick – ill”] 19
- 20. Even if KOSs try to capture and deliver the absolute meaning [or a more targeted one] of what they describe, they still are considered “collection independent knowledge structures” (Efthimiadis, 1996) ◦ There still are missing parts for the communication of the intended meaning Ambiguity differs only by degree between universal and domain-specific classifications, though that difference of degree is likely quite significant (Szostak, 2010) 20
- 21. Take advantage of *user models* Take advantage of *user evaluation* They deliver information about the real use of language The degree of ambiguity lessens within groups that regularly interact (though it does not disappear) (Szostak, 2010) 21
- 22. Thank you! Contact: Anna Mastora mastora [at] ionio [dot] gr This research has been co-financed by the European Union (European Social Fund – ESF) and Greek national funds through the Operational Program "Education and Lifelong Learning" of the National Strategic Reference Framework (NSRF) - Research Funding Program: Heracleitus II. Investing in knowledge society through the European Social Fund. 22