Difficulty-Gated Fusion of Reasoning Views for Temporal Retrieval

2026-08-09

AI Summary

This work addresses the challenge in temporal retrieval where query-document matching often fails due to reliance on temporal reasoning rather than lexical overlap, limiting the effectiveness of traditional equal-weight multi-view fusion. The authors propose an unsupervised, zero-shot, and reranking-free adaptive fusion method that constructs multiple query reformulation views explicitly encoding temporal intent. By leveraging an eight-dimensional feature vector—derived from score distributions, including softmax entropy, score gaps, and dispersion—the approach dynamically predicts query difficulty and employs a lightweight gating network to assign query-specific weights to each view. This is the first method to frame query performance prediction as a mechanism for determining multi-view fusion weights, enabling personalized view selection. Evaluated on the Tempo benchmark, it significantly improves performance across six retrieval architectures (p<0.001), achieving a peak nDCG@10 of 0.303 and yielding especially pronounced gains for weaker base models.

Abstract

Reasoning-intensive temporal retrieval requires matching a query to documents whose relevance depends on shared temporal reasoning rather than lexical overlap. Expanding a query into several reformulations that make its temporal intent explicit, and retrieving with each, supplies this reasoning, but fusing the resulting rankings with equal weights wastes accuracy: for any single query, only some reformulations are reliable. We propose query-difficulty-gated fusion of reasoning views. From each view we read an eight-dimensional signature of its score distribution, built from query-performance-prediction quantities such as softmax entropy, score gaps, and dispersion, and a gate of roughly one thousand parameters maps these signatures to per-query view weights. The fused ranking uses no relevance labels at inference, no re-ranking, and no fine-tuning of the retriever; the gate is trained leave-one-task-out. On the \textsc{Tempo} benchmark, the method improves all six retrievers we evaluate, from BERT encoders to 7B decoder retrievers, with the largest gains on the weaker backbones. The strongest retrievers reach $0.297$ and $0.303$ nDCG@10, and the per-query gain over the original query is significant under a paired bootstrap ($p<0.001$). A per-query oracle reaches $0.364$ against our realized $0.297$, exposing headroom that identifies per-query view selection as a concrete next step.

Problem

• temporal retrieval
• query reformulation
• reasoning views
• ranking fusion
• query difficulty

Innovation

• temporal retrieval
• query reformulation
• difficulty-gated fusion
• query performance prediction
• view weighting