CeQe: Grounding Lexical Retrieval in Semantic Evidence

2026-08-01

AI Summary

This work addresses the issue of relevant document omission in lexical retrieval methods like BM25, which stems from the semantic vocabulary gap. To mitigate this, the authors propose Cross-Encoder Query Expansion (CE-QE), a method that leverages token-level relevance attributions from cross-encoders applied to semantically retrieved documents to extract key terms for query expansion. By incorporating these terms into the original query without modifying the index, CE-QE enhances lexical recall while avoiding the self-reinforcing bias of traditional pseudo-relevance feedback and the hallucination risks of generative expansion. Integrated with BM25, semantic retrieval, and a novel score fusion strategy (SESF), the system achieves substantial performance gains across seven BEIR datasets—for instance, Recall@100 on NQ improves from 0.32 to 0.47. SESF further outperforms cross-encoder fusion by 2.5% in Recall@100 and surpasses SPLADEv2 and ColBERTv2 by 5.3% and 4.6% in nDCG@10, respectively.

Abstract

Lexical retrieval (BM25) captures exact keyword matches and weights terms by corpus-wide significance, but it is blind to the semantic vocabulary gap: when a relevant document phrases an answer differently from the query, BM25 never retrieves it, and no amount of downstream reranking or fusion can recover a document that was never in the candidate set. We present Cross-Encoder Query Expansion (CE-QE), which reads the per-token relevance attributions of a cross-encoder applied to top semantic search results, selects the terms the cross-encoder treats as decisive, and appends them to the BM25 query. Unlike classical pseudo-relevance feedback, which reuses BM25's own (possibly wrong) top results, CE-QE seeds expansion from the semantic retriever's results, avoiding self-reinforcing query drift. Unlike recent generative query expansion (HyDE, Query2doc), which prompts a large language model to hallucinate text from its parametric knowledge, every CE-QE expansion term is copied verbatim from a retrieved passage, so it cannot introduce vocabulary the corpus does not contain, and its only added cost is attribution extraction on a cross-encoder a hybrid pipeline already runs for reranking. On seven BEIR datasets, CE-QE improves lexical recall substantially where query and answer vocabulary diverge (e.g., NQ Recall@100 from 0.32 to 0.47), and its score-fusion variant (SESF) beats cross-encoder score fusion by 2.5% on Recall@100 and beats SPLADEv2 and ColBERTv2 by 5.3% and 4.6% on nDCG@10, while leaving the underlying BM25 index completely unmodified.

Problem

• lexical retrieval
• semantic vocabulary gap
• query expansion
• BM25
• retrieval recall

Innovation

• Cross-Encoder Query Expansion
• lexical retrieval
• semantic vocabulary gap
• query expansion
• relevance attribution