STEC: Evidence Compression for Deep Search in Open-domain Multi-Hop QA

2026-07-12

AI Summary

In open-domain multi-hop question answering, multiple search trajectories often yield heterogeneous, redundant, incomplete, or conflicting evidence, complicating answer selection. To address this challenge, this work proposes the STEC framework, which introduces a novel answer-level evidence compression mechanism. By normalizing search trajectories and grouping them according to candidate answers, STEC aggregates all trajectories supporting the same candidate into a unified, candidate-specific evidence representation. This compact representation then guides a verification process that selects answers based on comparative evidence quality at the candidate level, rather than directly comparing raw trajectories. This paradigm shift significantly enhances the reliability of answer selection. Experimental results demonstrate that STEC consistently outperforms existing methods across four mainstream benchmarks, and ablation studies confirm the effectiveness of evidence compression in improving answer selection accuracy.

Abstract

In open-domain multi-hop question answering (QA), LLM-based search agents offer a promising approach to knowledge-intensive QA by combining retrieval with reasoning. Existing methods mainly improve open-domain multi-hop QA through reasoning paradigms, retrieval interaction, and search strategy optimization. However, using multiple search trajectories introduces a challenging final answer selection problem. Different trajectories may support different candidates, and the retrieved information can be heterogeneous, redundant, incomplete, or conflicting. Directly comparing raw trajectories exposes the verifier to noisy and unaligned content, while comparing answer strings ignores the evidence supporting each candidate, making reliable final selection difficult. To address this challenge, we propose STEC, an evidence compression framework for final answer selection in multi-hop QA. STEC selects the final answer from the existing candidate set through two mechanisms: (1) Answer-Level Evidence Compression, which groups trajectories by normalized answer identity and converts each answer group into a candidate-specific evidence representation; and (2) Evidence-Guided Answer Verification, which compares these representations and selects the final answer from the candidate set. The design shifts final selection from raw trajectory comparison to candidate-level evidence comparison. We evaluate STEC on four open-domain multi-hop QA benchmarks against representative baselines. Experimental results show that STEC performs best overall among the compared methods, and ablation results provide evidence that answer-level evidence compression contributes to final answer selection.

Problem

• open-domain multi-hop QA
• final answer selection
• evidence compression
• search trajectories
• answer verification

Innovation

• evidence compression
• multi-hop QA
• answer selection
• search trajectories
• LLM-based search agents