AUTOMATIC QUERY REWEIGHTING USING CO-OCCURRENCE GRAPHS

Providing a relevant response to the user has always been challenging. Query reformulation methods have been widely applied in an attempt to provide a better representation of the user’s query and thus improve retrieval performance. In this paper, we present a new query reweighting method for document retrieval based on term co-occurrence graphs, which are built using a context window-based approach over the entire corpus. We propose an adapted version of the well-established Okapi BM25 model that allows identifying the most informative terms in the query and assigning them optimal weights. This measure stands out by its ability to evaluate the discriminative power of terms from co-occurrence graphs. Experimental results on two standard ad-hoc TREC collections show that our method improves both retrieval effectiveness and robustness and outperforms the state-of-the-art baselines with significant margins. 4, we can see that our method obtains a largely higher robustness score than the PRF methods in the Robust04 collection and achieves similar values in the Washington Post collection. These results confirm that our proposal is shown to be both more effective and more robust than the state-of-the-art baseline methods.