Jul 2026· Annual International ACM SIGIR Conference on Research and Development in Information Retrieval· 0 citations· 25 references
Computer Science
TL;DR
This work defines model-specific Token and Phrase Representativeness Scores (TRS/PRS) to discover tokens and key phrases that appear in a document but are poorly expressed by its embedding, which improve retrieval on a constructed hard-positive benchmark and can also improve standard in-domain retrieval benchmarks.
Abstract
Dense retrieval usually fails in two ways: ranking non-relevant documents too high, or ranking relevant documents too low. We focus on the second case from the query side: documents that are genuinely relevant but receive low retrieval scores under certain query formulations, forming hard-positive query–document pairs. To study this failure mode systematically, we build on recent token-alignment work, which analyzes vocabulary logits obtained by projecting retriever representations through an architecture-specific token prediction head and shows that low overlap among top-ranked logit tokens can cause relevant documents to be scored low. This suggests a practical route to hard positives: generate relevant queries that rely on document phrases that receive low ranks under these logits. We therefore define model-specific Token and Phrase Representativeness Scores (TRS/PRS) to discover tokens and key phrases that appear in a document but are poorly expressed by its embedding. Using high-PRS phrases as anchors, we automatically construct challenging yet relevant queries, yielding hard-positive query–document pairs. Experiments across multiple retrievers and datasets show that PRS-anchored queries induce substantially larger drops for dense retrievers than other query construction strategies, revealing a dense-specific weakness. Moreover, when mixed with standard hard negatives during contrastive fine-tuning, these hard positives improve retrieval on our constructed hard-positive benchmark and can also improve standard in-domain retrieval benchmarks. Our code is publicly available at https://github.com/wzy2001wzy/HardPositive.
The ad hoc document retrieval task is to rank documents by their presumed relevance to a query. Most TREC benchmarks provide relevance judgments only at the document level, without indicating which parts of a document are actually relevant. Focused relevance judgments, which highlight relevant text at the character level, are valuable but scarce. In this work, we present a study of ranking approaches, including lexical, dense and zero-shot prompted large language models (LLMs), to rank passages in relevant documents based on the presumed fraction of relevant text they contain. Our analysis shows that LLM-based rankings are highly effective and outperform strong sparse and dense retrieval baselines. We demonstrate the merits of our approaches in utilizing relevance feedback: constructing relevance models from top-ranked passages in relevant documents yields performance that transcends that of relevance models constructed from the entire documents.
Eyal El Ani, Eilon Sheetrit, O. Kurland· International Conference on...· 0 citations
Evaluation on WikiSA and ExaRank shows that ranking-based few-shot prompting generally improves over zero-shot prompting and achieves competitive performance against random-shot prompting, indicating that retrieval-based demonstration selection is beneficial but not uniformly superior in all settings.
A. Laksito, Aali Alqarni, Mark Stevenson· International Conference on...· 0 citations
Building question-answering systems that can read a document and answer naturally phrased questions about it is difficult when retrieval is left to either keyword matching or dense vector search alone, since each method has blind spots that surface as missed context, near-miss answers, or content invented by the underlying language model. This paper describes an optimized hybrid Retrieval-Augmented Generation (RAG) pipeline built to reduce these failure modes by combining two complementary retrieval signals: dense semantic similarity computed over a FAISS vector index, and sparse lexical scoring computed with BM25. Candidates returned by both retrievers are merged and passed through a Cross-Encoder re-ranking stage that scores each query-passage pair jointly, pushing the most contextually relevant chunks to the top before they reach the language model. Final answers are produced by Google’s Gemini model under a prompt that restricts it to the supplied context, which keeps the output tied to the source document rather than to whatever the model already “knows.” The pipeline is exposed through a Streamlit application that lets a user upload a PDF and ask questions about it in plain language, returning each answer alongside a confidence estimate and the page it came from. Evaluation on a multi-page technical PDF document shows that the hybrid retrieval and re-ranking stages together raise retrieval precision and reduce irrelevant or unsupported answers compared with retrieval limited to a single method, supporting the use of this approach for reliable, document-grounded question answering.
Vishwa K Dave, Pallavi· International Research Journ...· 0 citations
This work proposes Trident, with two complementary components: Trident-R, a retriever-agnostic LLM reranker that converts each candidate into an LLM-readable semantic record, then performs a single adaptive-K rerank call; and Trident-S, a generation-side module that prompts the VLM under topical, entity, and structural lenses before synthesis.
Guanchen Wu, Jiayuan Ding, Subhabrata Mukherjee et al.· 0 citations
LLM-based query expansion improves retrieval by generating document-like passages. In hybrid retrieval, however, most evaluations fuse fixed top-$L$ dense and sparse rankings. Because the cutoff controls both which cross-channel contributions enter fusion and how much of each ranking is accessed, gains measured at one $L$ can change or reverse at another. We separate these effects by evaluating retrieval effectiveness under complete-list fusion and recording the policy-specific per-channel replay stopping depths at which its ordered top-$K$ is certified. We then introduce DESA (Dense Expansion and Sparse Anchoring), a channel-asymmetric query expansion method. An LLM generates complementary reference passages; orthogonal residual expansion adds their new semantic directions to the dense query, while score-product anchoring incorporates their lexical cues into sparse retrieval without broadening the original query's lexical support. Across seven BEIR datasets, DESA improves nDCG@10 and Recall@20 over the unexpanded query by 3.82% and 2.38%, while reducing dense and sparse access depths by 36.90% and 36.56%. With equal dataset weighting, 63.31% of queries become shallower in both channels. However, both depths increase with Contriever on Touch\'e-2020. These results support channel-specific integration of generated passages and joint evaluation of retrieval effectiveness and access depth.
Lexical retrieval (BM25) captures exact keyword matches and weights terms by corpus-wide significance, but it is blind to the semantic vocabulary gap: when a relevant document phrases an answer differently from the query, BM25 never retrieves it, and no amount of downstream reranking or fusion can recover a document that was never in the candidate set. We present Cross-Encoder Query Expansion (CE-QE), which reads the per-token relevance attributions of a cross-encoder applied to top semantic search results, selects the terms the cross-encoder treats as decisive, and appends them to the BM25 query. Unlike classical pseudo-relevance feedback, which reuses BM25's own (possibly wrong) top results, CE-QE seeds expansion from the semantic retriever's results, avoiding self-reinforcing query drift. Unlike recent generative query expansion (HyDE, Query2doc), which prompts a large language model to hallucinate text from its parametric knowledge, every CE-QE expansion term is copied verbatim from a retrieved passage, so it cannot introduce vocabulary the corpus does not contain, and its only added cost is attribution extraction on a cross-encoder a hybrid pipeline already runs for reranking. On seven BEIR datasets, CE-QE improves lexical recall substantially where query and answer vocabulary diverge (e.g., NQ Recall@100 from 0.32 to 0.47), and its score-fusion variant (SESF) beats cross-encoder score fusion by 2.5% on Recall@100 and beats SPLADEv2 and ColBERTv2 by 5.3% and 4.6% on nDCG@10, while leaving the underlying BM25 index completely unmodified.
Adam Kahirov, Umesh Deshpande, S. Sundararaman· 0 citations