Negative feedback in the language modeling framework for text recommendation

Abstract

Text recommendation is the task of delivering sets of documents to users with respect to their profiles. One of the most important components of these systems is the filtering component. The filtering component decides about the relevancy of a document to a profile, which specifies the user interests, by comparing the similarity score between them with a predetermined threshold. In this paper, we propose a filtering approach which exploits the negative feedback from the user in a language modeling framework to compute the relevancy score of new documents. In other words, the negative feedback from the user is considered as the representative of the documents that he dislikes and leads the system to avoid suggesting such documents in the future. Our experiments on CLEF 2008–09 INFILE Track collection demonstrate the effectiveness of our proposed method and indicate that using negative feedback results in significant improvements over baselines.