Cluster-Based Fusion of Retrieved Lists

Methods for fusing document lists that were retrieved in response to a query often use retrieval scores (or ranks) of documents in the lists. We present a novel probabilistic fusion approach that utilizes an additional source of rich information, namely, inter-document similarities. Specifically, our model integrates information induced from clusters of similar documents created across the lists with that produced by some fusion method that relies on retrieval scores (ranks). Empirical evaluation shows that our approach is highly effective for fusion. For example, the performance of our model is consistently better than that of the standard (effective) fusion method that it integrates. The performance also transcends that of standard fusion of re-ranked lists, where list re- ranking is based on clusters created from documents in the list.