Hybrid document search method improves ranking on scientific fact dataset
Parameterized Dense-Sparse Fusion for Hybrid Retrieval: Tuning a Rank-Score Mix on BEIR SciFact with Qdrant
Information RetrievalArtificial Intelligence
Summary
Finding relevant documents can use two main methods: looking for exact word matches or checking if their meanings are similar. This paper studies a way to combine both methods using a few adjustable knobs to get better search results. The authors tested this on a scientific fact-checking dataset and found that tuning these knobs beats using either method alone. Their approach works well on this data but might need adjustment for other collections.
What this means in practice
- •For search platform engineers: Tune hybrid dense-sparse ranking parameters for improved scientific document retrieval quality on datasets like SciFact.
- •For enterprise search teams: Customize retrieval systems combining lexical and embedding search to boost recall and ranking on specialized text collections.
Authors
Satyanarayan Pati, Srikanth Patil
Abstract
We study a parameterized hybrid ranker that fuses a dense embedding list and a sparse lexical list. The method has a small, explicit parameter vector: a dense prior $α\in [0,1]$, a score-versus-rank mix $λ\in [0,1]$, an RRF smoothing parameter $κ> 0$, optional list-geometry coefficients that move $α$ per query, and a router margin $τ$ that can turn sparse search off. We grid-search those ranges on SciFact train (809 queries) and freeze the chosen values on SciFact test (300). The tuned rank-score mix ($α= 0.8$, $λ= 0.75$, $κ= 20$) reaches 0.753 nDCG@10 and 0.889 recall@10, outperforming dense BGE (0.742 / 0.871) and equal-weight RRF (0.707 nDCG@10) on that test split. A list-conditioned $α$ adds +0.0006 nDCG; a sparse-off router is rejected by the same train split (any $τ$ that skipped approximately 50% of queries lost nDCG). These coefficients are dataset-specific. Equal RRF with the same models does not beat dense on a nine-zip BEIR macro-average (0.479 vs. 0.519 nDCG@10). Repeating the same train-then-freeze sweep independently on all 20 indexed units beats equal RRF on 20/20 and dense on 16/20 (unit-mean nDCG@10 0.467 vs. 0.462 dense vs. 0.420 RRF). Other corpora should reuse the ranges, not a copy of the SciFact point.