通过分位数归一化统一图与向量检索得分，提升多跳问答的准确率。
Calibrated Fusion for Heterogeneous Graph-Vector Retrieval in Multi-Hop QA
- 用分位数排名法将不同分布的检索得分转为统一无量纲尺度。
- 在MuSiQue和2WikiMultiHopQA上，最后一步检索准确率分别提升至76.5%和53.6%。
- 该方法对后续融合方式不敏感，适合多源异构检索系统设计。
图增强检索结合密集相似度与基于图的相关性信号（如个性化页面排名PPR），但这些得分分布不同，难以直接比较。本文将此视为多跳问答中异构检索融合的得分校准问题。提出PhaseGraph方法，使用分位数排名归一化（PIT）将向量与图得分映射到统一无量纲尺度，实现稳定融合且保留幅度信息。在MuSiQue和2WikiMultiHopQA数据集上，校准融合使独立测试集上的最后一步检索性能提升：MuSiQue上LastHop@5从75.1%增至76.5%（8W/1L, p=0.039），2WikiMultiHopQA上从51.7%增至53.6%（11W/2L, p=0.023）。理论驱动的消融实验表明，分位数校准在调优与测试集上均比最小-最大归一化更具方向鲁棒性（1W/6L, p=0.125），而校准后布兹曼加权与线性融合表现相当（0W/3L, p=0.25）。结果表明，得分共度是稳健的设计选择，具体校准后融合算子影响较小。
原文摘要 · Abstract (English)
Graph-augmented retrieval combines dense similarity with graph-based relevance signals such as Personalized PageRank (PPR), but these scores have different distributions and are not directly comparable. We study this as a score calibration problem for heterogeneous retrieval fusion in multi-hop question answering. Our method, PhaseGraph, maps vector and graph scores to a common unit-free scale using percentile-rank normalization (PIT) before fusion, enabling stable combination without discarding magnitude information. Across MuSiQue and 2WikiMultiHopQA, calibrated fusion improves held-out last-hop retrieval on HippoRAG2-style benchmarks: LastHop@5 increases from 75.1% to 76.5% on MuSiQue (8W/1L, p=0.039) and from 51.7% to 53.6% on 2WikiMultiHopQA (11W/2L, p=0.023), both on independent held-out test splits. A theory-driven ablation shows that percentile-based calibration is directionally more robust than min-max normalization on both tune and test splits (1W/6L, p=0.125), while Boltzmann weighting performs comparably to linear fusion after calibration (0W/3L, p=0.25). These results suggest that score commensuration is a robust design choice, and the exact post-calibration operator appears to matter less on these benchmarks.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。