An experimental comparison of relevance-feedback techniques

Abstract

This paper describes an experiment in which seven relevance feedback document retrieval techniques are compared. It is shown that for the set of parameters and indexed document collection used, two of the techniques performed worse than the initial queries and that the greatest gains in precision for the others occur at low levels of recall. At the extreme recall levels no significant differences were found among any of the techniques while over all recall levels no significant differences were found among five of the techniques.