Incremental profile learning based on a reinforcement method

Abstract

This paper presents a profile learning method in an adaptive information filtering. The learning method is an incremental profile learning based on a reinforcement algorithm. The basic idea consists in building, when a document is selected and judged as relevant, the temporary profile which makes it possible to find this document with a strong score, then integrating this profile, using a logarithmic function, in the global profile. The proposed method is compared to two IR learning methods, query expansion method used in Okapi and Rocchio's algorithm. Experiments carried out on TREC1-2002 collection showed the effectiveness of the reinforcement method.Adaptive filtering, profile learning, reinforcement method