We introduce SciRet, a compute-aware empirical study of retrieval-augmented generation for scientific question answering over CORD-19. Rather than proposing a new model, we evaluate a fixed scientific RAG pipeline across three corpus scales: 1,034 chunks (1K papers), 5,160 chunks (5K papers), and 15,480 chunks (15K papers). The pipeline combines sentence-window chunking, BM25, BGE-M3 dense retrieval, reciprocal rank fusion, optional cross-encoder reranking, and grounded answer generation. Across these settings, hybrid retrieval is more robust than either sparse-only or dense-only retrieval in our setting, reaching Recall@10 of 1.000 at 1K and 15K. In contrast, an MS MARCO-trained cross-encoder reranker reduces precision on the scientific corpus, suggesting that domain mismatch can outweigh the benefits of stronger query-passage interaction. Generation faithfulness measured with RAGAS increases with corpus scale in our setup. Retrieval evaluation uses pseudo-relevance labels derived from the hybrid system, so we treat the results as controlled comparative evidence rather than a benchmark claim. We release code, indexes, and evaluation outputs to support replication and follow-up studies.
Retrieval-Augmented Generation (RAG) is the standard for grounding Large Language Models (LLMs), but its parts' embeddings, lexical scoring, query expansion, reranking, and adaptive sizing do not always compose additively. Under default configurations, adding naïve hybrid retrieval with Reciprocal Rank Fusion degrades NDCG@10 by up to 5.6% on scientific corpora; temperature-sampled query expansion with a small LLM further degrades it by 16.8%. To address this, we propose the Adaptive Multi-Stage Vector Retrieval (AMSVR) framework, prioritising weighted, drift-resistant composition over uniform fusion. Rather than a one-sizefits-all system, AMSVR offers tailored configurations: AMSVR-Scientific (dense + tuned hybrid) peaks at NDCG@10 = 0.7570 on SciFact, while AMSVR-Full (seven stages) targets broader, noisier corpora where Recall@100 matters most. We evaluate on three BEIR benchmarks (SciFact, NFCorpus, FIQA), release an offline configuration-diagnostic tool, and provide a per-corpus recipe for selecting which stages to enable.
Samsudeen Alabi Bankole, Yakub Kayode Saheed· NLP & Big Data· 0 citations
We present a trust-weighted Retrieval-Augmented Generation (RAG; Lewis et al., 2020) system for SemEval-2026 Task 7 (BLEnD) Track 2 (Ousidhoum et al., 2026), targeting English cultural multiple-choice QA across 30 countries. Built atop Llama-3.1-8B-Instruct (Meta AI, 2024), the six-phase pipeline integrates hybrid BM25+FAISS retrieval, country-aware filtering, intent detection, tiered routing, anti-leak prompt engineering, and trust-weighted reranking. The core finding is that RAG hurts rather than helps: the LLM-only baseline achieves 78.6% accuracy, outperforming the full system at 78.5% (McNemar’s test, p = 0 . 962 ). Oracle analysis reveals that only 40.7% of questions are answerable from the knowledge base, explaining why retrieval introduces more noise than signal. The sole recovery comes from anti-leak prompt filtering (Phase 4), which mitigates answer-anchoring artifacts. Code: https://github.com/CultRAG/ BLEnD-CultRAG .
This system paper presents AMU’s submission to RAG4Reports 2026 Task B: a practical multilingual retrieval-augmented generation pipeline for evidence-supported report generation. The system combines full-query retrieval, optional query rewriting, dense retrieval with Qdrant, cross-encoder reranking, diversity-aware context selection, and structured generation. The best submitted run uses BAAI/bge-m3 embeddings, BAAI/bge-reranker-v2-m3 reranking, and gpt-5.1 generation with medium reasoning effort, using a partial-coverage prompting strategy. On the official leaderboard, it achieved F1 =0.4351, sentence_support =0.8280, and nugget_coverage =0.3403, indicating that the generated reports were well grounded but only partially comprehensive.
Maciej Czajka, Piotr Jabłoński, Mateusz Czajka et al.· Proceedings of the 1st Works...· 0 citations
This paper presents the IIMAS-RAG system submitted to SemEval-2026 Task 8, which evaluates multi-turn retrieval-augmented generation (RAG) conversations. Our system is a modular pipeline composed of three stages: (1) LLM-based query rewriting to transform conversational history into standalone queries, (2) hybrid sparse–dense retrieval combining SPLADE and Voyage-3-large via Reciprocal Rank Fusion (RRF), and (3) answerability-conditioned generation using GPT-4.1. In Sub-task A (Retrieval), our system ranked 4th out of 38 teams (nDCG@5 = 0.5445), demonstrating the robustness of the hybrid retrieval strategy in specialized domains. On Subtask C (Full RAG), we ranked 13th out of 29 teams (composite = 0.5397). Ablation experiments show that LLM-based query rewriting is the main driver of retrieval performance, yielding a +16.3% relative gain in nDCG@10 over the hybrid baseline without rewriting, while domain-specific prompt variants provide only localized gains on specialized corpora. Generative performance remains sensitive to low-context and partially answerable turns, where the user query lacks sufficient grounding information and the model struggles to either abstain or provide a properly qualified partial answer, explaining the performance gap between retrieval and final synthesis. Our code is available at https://github.com/PLN-disca-iimas/ mtrag_semeval2026 .
Vania Raya-Rios, Helena Gómez-Adorno, Leon Hecht et al.· SemEval@ACL· 1 citation
Retrieval-augmented generation (RAG) systems depend on retrieval modules to supply grounding evidence for large language models. While hybrid approaches combining sparse and dense retrievers improve performance, most rely on fixed weights that ignore query-specific and corpus-specific variation. Similarly, query expansion has long been used to enrich recall, but its integration with original queries is usually static and can introduce noise. We present Q U DAR, a dual-perspective adaptive retrieval framework motivated by a systematic analysis of retrieval behavior across retriever type (sparse vs. dense) and query format (original vs. expanded). Leveraging margin-derived confidence (e.g., top-1–top-2 score gaps) and LLM-based relevance scoring, Q U DAR dynamically assigns query-specific weights, enabling effective integration of complementary retrieval signals while mitigating noise. Q U DAR is lightweight, retriever-agnostic, and broadly applicable. Experiments show consistent gains over static baselines, improving retrieval quality by 12 – 16 % and yielding more stable performance across queries.
Joeun Kim, Seunghyouk Yoon, Xuan-Bach Le et al.· Annual Meeting of the Associ...· 1 citation
Long-document visual question answering (VQA) over documents of tens to hundreds of pages mixing text, tables, charts, and figures typically follows retrieve-then-read pipelines. In our setting, the bottleneck shifts from retrieval recall to reranker-side evidence selection: on MMLongBench-Doc, BGE-M3 reaches Recall@20 = 0.86 but only F1@5 = 0.254, and even the visual retriever ColPali reaches only F1@5 = 0.332; a text-only rerank LLM seeing only raw snippets misses table, chart, and layout evidence even when the upstream retriever encoded images. We propose Trident, with two complementary components: Trident-R, a retriever-agnostic LLM reranker that converts each candidate into an LLM-readable semantic record, including a visual caption, section path, entity tags, multi-axis concept hits, and a text snippet, 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. On two long-document datasets, the annotation+rerank protocol substantially improves retrieval F1 across five heterogeneous pools, with every reranked pool exceeding the strongest adaptive-K baseline PageIndex. An LLM rerank without the annotation barely changes first-hit ranking, indicating the lift comes from the structured annotation. Trident-S targets open-ended synthesis questions by design, adding up to 6.6 points in generation accuracy on these questions. The best Trident configuration is the strongest downstream QA pipeline in our evaluation, with rankings consistent across two LLM judges (kappa = 0.913).
Guanchen Wu, Jiayuan Ding, Subhabrata Mukherjee et al.· 0 citations