|
Word-sense Disambiguation in the Absence of
Textual Context Using Yahoo-derived Models of User Interests
Rosie Jones
Computational Linguistics Program and Text-Learning Group,
Carnegie-Mellon University
When: Tuesday, 25th March 1997
Time: 10:00am
Where: Room E6A357, Macquarie University
Abstract:
Word-sense disambiguation is a a well-researched problem in natural language processing, since it causes difficulties for translation systems, understanding systems, and information retrieval systems. Common statistical techniques involve using the surrounding textual context to disambiguate the word. Information Retrieval evaluations frequently rely on queries of sufficient length that word-context will aid in disambiguation of certain terms. However, typical lay-person search queries (such as to web-based search engines) are on average 1.7 words long - too few to use any large or small textual context window for disambiguation. Knowledge of other aspects of the context can aid in disambiguation. In particular, modelling the user's long-term interests, based on a history of web accesses to text documents, can aid in individualised disambiguation of words for search queries.
Enquiries: sals@mri.mq.edu.au
| Last modified: July 1997 |