An Analysis of Fusion Functions for Hybrid Retrieval

We study hybrid search in text retrieval where lexical and semantic are fused together with the intuition that two complementary how they model relevance. In particular, we examine fusion by a convex combination (CC) of scores, as well Reciprocal Rank Fusion (RRF) method, identify their advantages potential pitfalls. Contrary to existing studies, find RRF be sensitive its parameters; learning CC is generally agnostic choice score normalization; outperforms in-domain out-of-domain settings; finally, sample efficient, requiring only small set training examples tune parameter target domain.