Association Is Not Similarity: Learning Corpus-Specific Associations for Multi-Hop Retrieval

TL;DR

The paper introduces Association-Augmented Retrieval (AAR), a lightweight method that improves multi-hop passage retrieval by learning corpus-specific associations, significantly enhancing recall on benchmarks.

Contribution

AAR is a novel, efficient reranking approach that captures corpus-specific associations for better multi-hop retrieval without requiring large language models.

Findings

AAR improves HotpotQA Passage Recall@5 from 0.831 to 0.916.

AAR achieves +10.1 points on MuSiQue in the transductive setting.

Retrieval gains translate to +6.4 exact match in downstream QA.

Abstract

Dense retrieval systems rank passages by embedding similarity to a query, but multi-hop questions require passages that are associatively related through shared reasoning chains. We introduce Association-Augmented Retrieval (AAR), a lightweight transductive reranking method that trains a small MLP (4.2M parameters) to learn associative relationships between passages in embedding space using contrastive learning on co-occurrence annotations. At inference time, AAR reranks an initial dense retrieval candidate set using bi-directional association scoring. On HotpotQA, AAR improves passage Recall@5 from 0.831 to 0.916 (+8.6 points) without evaluation-set tuning, with gains concentrated on hard questions where the dense baseline fails (+28.5 points). On MuSiQue, AAR achieves +10.1 points in the transductive setting. An inductive model trained on training-split associations and evaluated on unseen validation associations shows no significant improvement, suggesting that the method captures corpus-specific co-occurrences rather than transferable patterns. Ablation studies support this interpretation: training on semantically similar but non-associated passage pairs degrades retrieval below the baseline, while shuffling association pairs causes severe degradation. A downstream QA evaluation shows retrieval gains translate to +6.4 exact match improvement. The method adds 3.7ms per query, trains in under two minutes on a single GPU, and requires no LLM-based indexing.