bm25 retrieval

This paper addresses the high latency and computational overhead of Learning-to-Retrieve Sparse (LSR) methods in long-document retrieval, systematically investigating their reproducibility and adaptation mechanisms. We propose a segment-aggregation-based LSR optimization framework; ablation studies reveal that the first segment dominates retrieval effectiveness and underscore the critical role of global context modeling for long-document adaptation. Building upon ExactSDM and SoftSDM, we introduce several post-processing enhancements—n-gram augmentation, proximity-weighted scoring, representation fine-tuning, and multi-signal fusion—to significantly improve paragraph-level relevance modeling. Experiments demonstrate both effectiveness and reproducibility, and uncover a consistent positional sensitivity pattern: early segments exert disproportionate influence on final retrieval scores. All code and comprehensive experimental results are publicly released.

To address the limitation of coarse-grained embeddings in modeling fine-grained semantic details for long-document retrieval, this paper proposes a fine-grained chunked embedding and weighted relevance aggregation framework. First, documents are segmented into semantically coherent paragraph-level chunks, and large language models generate chunk-level embeddings. Second, chunk–query relevance scores are computed and aggregated via a learnable weighting mechanism to produce precise document-level representations. Third, an optimized pairwise ranking loss is introduced to enhance retrieval ranking performance. Evaluated on multiple long-document retrieval benchmarks, the method significantly outperforms single-document embedding baselines, achieving substantial gains in retrieval accuracy while reducing embedding generation latency by over 40%. The approach thus achieves a favorable trade-off between efficiency and effectiveness.

To address the semantic gap in long-document semantic retrieval—where conventional keyword-based methods neglect lexical semantics—this paper proposes a dual-representation vector space model integrating lexical and semantic features. Our method introduces three key innovations: (1) a semantic word clustering algorithm grounded in pre-trained word embeddings to automatically identify semantically cohesive term groups; (2) a cluster-level weighting scheme that fuses intra-cluster term frequency with an enhanced TF-IDF variant; and (3) a joint lexical-semantic matching function that preserves exact lexical matching capability while substantially improving semantic coverage. Experimental evaluation on SQuAD and TREC-CAR demonstrates statistically significant improvements over pure keyword baselines: semantic retrieval accuracy increases markedly, without degrading traditional lexical retrieval performance. The approach thus achieves a balanced trade-off between precision and semantic robustness in long-document retrieval.

Existing hybrid search systems lack systematic empirical analysis of trade-offs among lexical and semantic retrieval components—i.e., retrieval paradigms, fusion strategies, and re-ranking methods—leading to complex, suboptimal configurations. Method: We introduce the first benchmark framework tailored for advanced hybrid architectures, conducting systematic evaluation across 11 real-world datasets, covering four retrieval paradigms, their combinations, and re-ranking strategies. Contribution/Results: We identify a “weakest-link” effect in hybrid pipelines and propose a data-driven configuration mapping method. Crucially, we find Tensor-based Re-ranking Fusion (TRF) achieves both high efficiency and strong semantic modeling under low-resource conditions, overcoming traditional fusion bottlenecks. Experiments reveal that hybrid performance is severely constrained by imbalanced path quality; optimal configurations are highly dependent on dataset characteristics and resource constraints. TRF significantly improves the effectiveness–cost trade-off, outperforming state-of-the-art baselines across diverse settings.

This work addresses the limitations of single-retrieval approaches in semantic and keyword matching for large language model agents operating in local-first settings. We propose a hybrid retrieval framework that constructs a local document memory system using SQLite, integrating vector similarity search (via sqlite-vec) with full-text keyword matching (FTS5). The system employs an adaptive Reciprocal Rank Fusion (RRF) strategy combined with per-query IDF weighting. Innovatively, we introduce a self-supervised embedding fine-tuning method based on retrieval disagreement, optimizing BGE-small with MultipleNegativesRankingLoss. The framework also incorporates integrity verification, version control, and diagnostic tools. Experiments demonstrate up to a 19.5% improvement in NDCG@10 across five BEIR datasets, with our 33M-parameter model matching or surpassing the 110M-parameter ColBERTv2 and BGE-base on three datasets, while achieving a median retrieval latency of only 20.9ms over 50,000 text chunks.

This study investigates whether retrieval fusion techniques—commonly adopted in real-world retrieval-augmented generation (RAG) systems, such as multi-query retrieval and reciprocal rank fusion—consistently improve end-to-end answer quality under practical deployment constraints. Conducted within an enterprise knowledge-base RAG pipeline, the evaluation is performed under fixed retrieval depth, reranking budget, and latency limits. While retrieval fusion enhances initial recall, it fails to translate into improved Top-k accuracy after subsequent reranking and context truncation; notably, Hit@10 declines from 0.51 to 0.48 and incurs additional latency. These findings challenge the prevailing assumption of the default efficacy of recall-oriented fusion strategies, revealing diminishing returns in production settings where downstream processing and system constraints critically shape overall performance.

This study addresses the lack of systematic evaluation of modern retrieval methods in Retrieval-Augmented Generation (RAG) systems for heterogeneous financial documents containing both text and tables. It presents the first large-scale benchmark comparing ten retrieval strategies—including sparse, dense, hybrid fusion, cross-encoder reranking, and query expansion—on a financial mixed-document question answering task. Evaluation employs multiple metrics (Recall@k, MRR, nDCG, and numerical match accuracy) alongside bootstrap significance testing. Results reveal that BM25 outperforms state-of-the-art dense retrievers in this domain; a two-stage hybrid retrieval pipeline with neural reranking achieves substantial gains (Recall@5 = 0.816, MRR@3 = 0.605); context-aware retrieval consistently improves performance, whereas query expansion yields limited benefits for numerical questions. The authors release a comprehensive open-source benchmark and provide practical guidance on cost–accuracy trade-offs.

This work addresses the limitation of traditional retrieval systems, which treat document representation as a static preprocessing step and thus struggle to adapt to downstream tasks. The authors propose AutoIndex, a novel framework that formulates document representation construction as a learnable program synthesis problem. AutoIndex dynamically generates retrieval-oriented representations by searching over executable transformation programs—such as slicing, augmentation, and normalization—and iteratively refines them using validation feedback. By integrating proxy-guided program search with retrieval quality evaluation, the method enables explicit optimization of document representations. Evaluated on the CRUMB benchmark across all eight tasks, AutoIndex consistently outperforms the full-document BM25 baseline, achieving average improvements of 8.4% in Recall@100 and 8.3% in nDCG@10, with peak gains reaching 30.5% and 43.6%, respectively.

This work addresses the limitations of conventional RAG systems in high-precision citation scenarios, where flat chunking leads to structural information loss, lexical mismatches, and unstable reasoning, thereby failing to meet the accuracy and traceability demands of technical question answering. To overcome these challenges, the authors propose a hierarchical dense retrieval framework that constructs a four-level tree-structured document index (document → section → paragraph → sentence) and employs bottom-up embedding aggregation to preserve hierarchical context. The approach further integrates LLM-driven query planning and cross-query reranking to enhance retrieval coverage, alongside an ensemble reasoning mechanism with an abstention protocol to stabilize outputs. Remarkably, using dense retrieval alone, this method matches the performance of sparse-dense hybrid strategies, achieving a score of 0.861 to rank first on both public and private leaderboards in the WattBot 2025 Challenge—the only team to top both evaluation tracks. Ablation studies confirm the contribution of each component, and the code is publicly released.