DualView: Adaptive Local-Global Fusion for Multi-Hop Document Reranking

TL;DR

DualView introduces an efficient adaptive local-global fusion framework for multi-hop document reranking, significantly improving recall and accuracy while reducing latency compared to larger models.

Contribution

It proposes a novel dual-view cascaded reranking architecture with adaptive fusion, enhancing multi-hop document retrieval efficiency and effectiveness.

Findings

Achieves 99.4% Top-4 Recall on MuSiQue with 4.0 ms latency

Outperforms larger cross-encoders in recall metrics

Both local and global views are crucial for performance

Abstract

Multi-hop question answering requires aggregating information from multiple documents, a critical capability for knowledge-intensive applications. A fundamental challenge lies in efficiently identifying the minimal relevant document set from retrieved candidates while maintaining high recall. We present an efficient dual-view cascaded reranking framework for multi-hop document reranking. Operating as a lightweight post-retrieval stage over E5-base-v2 candidates, our architecture comprises: (1) a Local Scorer employing stacked cross-attention for fine-grained query-document relevance; and (2) a Global Scorer modeling inter-document dependencies via Transformer-based context aggregation. These views are dynamically fused through an Adaptive Gate conditioned on query semantics. Under the fixed candidate set reranking setting with offline cached embeddings, our model achieves competitive results, particularly outstanding on MuSiQue with 99.4% Top-4 Recall and 97.8% Full Hit accuracy at 4.0 ms latency (249 QPS). It substantially outperforms 600M-parameter cross-encoders (BGE-Large: 92.0% Recall, Jina-v3: 90.1% Recall) while maintaining 5 to 6 times lower latency. Ablation studies validate that both Local and Global views contribute substantially to multi-hop performance.