Improving document retrieval with a clustering based relevance feedback system

Relevance feedback for document retrieval systems is a technique where user feedback is used to improve a query response. In this work we propose a system that uses multiple clusterings and a semi-supervised heuristic to improve a query response. The heuristic creates an optimal cluster w.r.t. the relevance feedback based on multiple clusterings. We justify the explicit separation of the optimization process and the clustering process by time and space constrains. The evaluation of the heuristic on a corpus containing 1.660 documents from MEDLINE showed promising results. We were able to obtain better results as a single clustering after a few iterations