Experiments on pseudo relevance feedback using graph random walks

Abstract

In this article, we apply a graph-based approach for pseudo-relevance feedback. We model term co-occurrences in a fixed window or at the document level as a graph and apply a random walk algorithm to select expansion terms. Evaluation of the proposed approach on several standard TREC and CLEF collections including the recent TREC-Microblog dataset show that this approach is in line with state-of-the-art pseudo-relevance feedback models. © 2012 Springer-Verlag Berlin Heidelberg.