Capturing Information Need by Learning User Context

Learning techniques can be applied to help information retrieval systems adapt to users' spe-ciic needs. They can be used to learn from user searches to improve subsequent searches. This paper describes the approach taken to learn about particular users' contexts in order to improve document ranking produced by a probabilistic information retrieval system. The approach is based on the argument that there is a pattern in user queries in that they tend to remain within a particular context over on-line sessions. This context, if approximated, can improve system performance. While it is not uncommon to link the evidence from one query to the next within a particular online session, the approach here groups the evidence over diierent sessions. The paper concentrates on the user-oriented evaluation method used in order to determine whether or not the approach improved information retrieval system performance .