Concepts and Context for Information Retrieval
Abstract
Most people are now very familiar with word-based statistical information
retrieval through their use of web search engines. Search failures with
these engines are often blamed on their reliance on words rather than "concepts."
This talk will describe the differences between these two representations
from an IR perspective, with an emphasis on the importance of context or
word co-occurrence. Examples of research at the CIIR that exploits context
will be given, including corpus-based stemming, query expansion, phrase
representations, cross-lingual retrieval, and analysis of "core concepts."
Luis Gravano
gravano@cs.columbia.edu