DualView：面向多跳文档重排序的自适应局部-全局融合框架（已撤稿）

This paper has been withdrawn by Litong Zhang

No PDF available, click to view other formats

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.

Comments:	This version has a critical dataset error in Section 4.1.2. The MuSiQue benchmark used is not the original dataset but a modified variant incorporating hard negatives from HotpotQA and MultiWiki, causing data distribution shifts. Thus, comparisons in Section 4.2 are invalid due to non-standard evaluation
Subjects:	Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
Cite as:	arXiv:2605.18767 [cs.IR]
	(or arXiv:2605.18767v2 [cs.IR] for this version)
	https://doi.org/10.48550/arXiv.2605.18767 arXiv-issued DOI via DataCite

Submission history

From: Litong Zhang [view email]
[v1] Mon, 13 Apr 2026 08:56:09 UTC (7,836 KB)
[v2] Wed, 21 Oct 2026 04:30:16 UTC (1 KB) (withdrawn)