Context learning in Okapi

Abstract

A user who makes repeated use of a retrieval system may be assumed to have a context which is common to successive uses (even if the immediate need differs). An IR system which could make use of this context may be better able to match the specific need. A machine-learning approach to inferring the user's context is described, and the results of an evaluation experiment are given. There appears to be scope for IR systems to operate in this way.