Evidence-Unit Fairness and the Limits of Query-Adaptive Sparse-Dense Fusion in Financial Document Retrieval

2026-07-31

AI Summary

This study addresses the challenges of financial document retrieval, where queries are often short and contain abbreviations, and relevant evidence is scattered across lengthy documents and tables, limiting the effectiveness of conventional sparse and dense retrieval methods. To overcome these issues, the authors segment documents into passages aligned with the encoder’s input window to fully cover annotated evidence spans, combine BM25 with a compact dense retriever, and apply Reciprocal Rank Fusion (RRF) to enhance performance. The work further reveals that passage length introduces bias in dense retriever evaluation, demonstrates that training-free RRF outperforms uniform-weight fusion, and investigates three lightweight adaptive routing strategies. On the revised FinDER dataset, fixed-weight fusion improves Hit@10 by 28%; however, despite a theoretical improvement margin of 21.8%, the adaptive approaches fail to achieve statistically significant gains.

Abstract

Retrieval over financial filings is difficult because queries are short and acronym-heavy while the answer-bearing evidence sits inside long, table-dense documents. We study sparse-dense hybrid retrieval on FinDER, a benchmark of expert-annotated questions over corporate 10-K filings. Our first finding is methodological: if the retrieval unit is larger than the dense encoder's input window, the dense model never sees a large share of the labeled evidence, confounding comparison against a full-text sparse baseline. We measure this directly and remove it by segmenting the corpus into encoder-sized windows. On the corrected corpus, fusing BM25 and a compact dense encoder improves reference-level Hit@10 by roughly 28 percent over either component, and training-free, untuned reciprocal rank fusion exceeds the equal-weight blend in an exploratory comparison. We then ask whether choosing the fusion weight per query helps: an oracle over the interpolation-weight grid shows headroom of 21.8 percent, yet none of the three lightweight adaptive routers (a score-confidence heuristic, a random forest over query features, and a ridge regressor over query embeddings) establishes a statistically reliable improvement over the fixed blend under company-grouped cross-validation with cluster-robust inference. Simple fusion is a strong baseline here, and we discuss why per-query weighting does not capture the available headroom.

Problem

• financial document retrieval
• sparse-dense fusion
• evidence-unit fairness
• query-adaptive weighting
• retrieval evaluation

Innovation

• evidence-unit fairness
• sparse-dense fusion
• query-adaptive retrieval
• financial document retrieval
• reciprocal rank fusion