We describe the participation of the University of Amsterdam’s ILPS group in the relevance feedback track at TREC 2008. We introduce a new model which incorporates information from relevant and non-relevant documents to improve the estimation of query models. Our main findings are twofold: (i) in terms of statMAP, a larger number of judged non-relevant documents improves retrieval effectiveness and (ii) on the TREC Ter- abyte topics, we can effectively replace the estimates on the judged non-relevant documents with estimations on the document collection.

[bibtex key=TREC:2009:meij]

http://www.slideshare.net/edgar.meij/query-modeling-using-nonrelevance-information-trec-2008-relevance-feedback-track-talk-edgar-meij-univ-of-amsterdam-presentation