Jul 2026· Annual International ACM SIGIR Conference on Research and Development in Information Retrieval· 0 citations· 18 references
Computer Science
TL;DR
These findings highlight numerical fabrication as a critical gap in current hallucination detection approaches and recommend the need for specialized, number-aware methods in RAG systems.
Abstract
With the rise in the usage of Retrieval-Augmented Generation (RAG) systems to improve the factual accuracy of the large language models (LLM), there still exists a concern regarding these systems producing hallucinating outputs not grounded in the retrieved documents. Although prior work has studied general hallucination detection, the specific challenge of numerical fabrication remains unquantified. This research study analyzes 500 RAG outputs using GPT-3.5-turbo on MS MARCO queries to address this specific challenge and found that 38.2% of failures involve numerical fabrication. The evaluation consists of four detection methods that span different paradigms: embedding-based (Semantic Similarity), metric-based (BERTScore), LLM-based (GPT-4o-mini), and a number-aware heuristic. The results show that all standard methods struggle with numerical hallucinations, notably GPT-4o-mini achieves only 25.7% recall on numerical failures despite being a state-of-the-art LLM judge. In contrast, the simple number-aware heuristic of this research study achieves a 100% recall on numerical failures with F1=0.616, significantly outperforming all baselines (McNemar's test, p<0.001). These findings highlight numerical fabrication as a critical gap in current hallucination detection approaches and recommend the need for specialized, number-aware methods in RAG systems.
The issue of hallucination detection for Retrieval-Augmented Generation (RAG) systems is an ongoing reliability challenge that arises with the scale up of LLM deployment. We introduce the HaloGuard: a systems-integration study that integrates three types of signals, multi-judge LLM consensus, uncertainty quantification through temperature sampling, and lexical text features, and evaluate with LibreEval (10,871 samples; GPT-4o, Claude-3.5-Sonnet, Llama-3.1-8B). They all rely on previous work; the novel aspect of the present paper is their principled combination, empirical decomposition through ablation, calibration analysis, and deployment characterization over operating thresholds, hallucination types, and six languages At a conservative threshold, HaloGuard achieves AUC-ROC = 0.986 [95% CI: 0.981-0.991] and F1 = 0.665 [0.639-0.690]; at the balanced max-F1 threshold, F1 = 0.769 (Precision = 0.812 / Recall = 0.731), directly addressing the precision-1.000 operating-point artifact. A fine-tuned variant (HaloGuard-FT, DeBERTa-v3-base) reaches F1 = 0.847 [0.831-0.863] and AUC = 0.994 [0.991-0.997]. Weight sensitivity analysis confirms F1 is stable across judge weight alpha in [0.5, 0.7] (F1 range <= 0.016), ruling out weight cherry-picking. Additional analyses quantify a 2.15x novelty differential (p < 0.001), source-attribution correlation (r = -0.319), and systematic cross-lingual F1 degradation (EN: 0.665 -> JA: 0.543). Calibration is well-behaved (ECE = 0.043). All results are scoped to LibreEval; cross-dataset validation on RAGTruth is the stated priority next step. Full judge prompts, hyperparameters, and code are documented for reproducibility.
Harshath R, G. S, S. Thangavel et al.· 2026 International Conferenc...· 0 citations
This work analyzes three small-to-mid-sized open-weight LLMs in combination with multiple retrieval strategies, including sparse, dense, and hybrid approaches, as well as score-based and rank-based fusion techniques, enabling a comprehensive assessment of retrieval-generation interactions.
C. Mala, Gizem Gezici, Fosca Giannotti· Machine-mediated learning· 0 citations
Although the performance of LLMs on a wide variety of natural language processing problems demonstrates remarkable ability, hallucinated responses are introduced as one of the key weaknesses of LLMs, especially in knowledge-intensive applications, where fidelity to facts is paramount. While effective in reducing hallucinations, the context retrieved during the RAG operations is still very often suboptimal with respect to the external documents used in the retrieval stage, and being based on cosine similarity and nearest-neighbour search, these models typically do not return optimal context to support factual generation. The authors propose a five-stage solution, called Quantum-Enhanced Retrieval-Augmented Generation (QERAG), which combines the use of a quantum-inspired probability amplitude document ranking, a Context Utility Score (CUS) optimisation engine, and a Hallucination Verification Agent (HVA) based on a formal Hallucination Reduction Index (HRI). The Quantum Relevance Score (QRS) uses interference based, Amplitude Encoding on candidate Document Sets to yield a normalised relevance distribution that emphasises the relatively higher value generated in the discriminative context as compared to other locally similar passages in a document set. QERAG outperforms Standard LLM and Traditional RAG models on all five Natural benchmark across response accuracy, with a score of 92.8% vs. 74.3%, and hallucination rate, 4.3% vs. 18.2%, respectively. They achieve a significant overall improvement by using quantum-inspired retrieval layer, which achieves a 3.5-percentage-point reduction in hallucination rate when compared to cosine-similarity RAG, with this single layer showing more reduction than the rest of the pipeline layers. The results of an ablation study show the individual contribution of every pipeline stage, and the quantum-inspired retrieval layer exhibits the greatest reduction of a 3.5 percentage points over cosine-similarity RAG.
Praveenkumar Seepana· International Journal of Com...· 0 citations
Retrieval-Augmented Generation (RAG) has been increasingly adopted to reduce hallucinations and strengthen the factual grounding of large language models (LLMs). While robustness to errors in the retrieval process has been explored, the impact of ideological bias on LLM outputs has been overlooked. For instance, if the retrieved material contains ideological positions, the RAG may transmit, amplify, or suppress such ideological discourses in its outputs. In this study, we address this issue by examining the influence of the RAG framework, comprising ideological discourses, in LLM-generated answers. To this end, we applied Lexical Multidimensional Analysis (LMDA) on a corpus of 1,117 COVID-19 treatment articles, identifying three ideological discourses. This corpus is then used as the external knowledge source for the RAG. We assessed several LLMs by having the models answer ideological questions at different sampling temperatures. The generated texts were assessed semantically and lexically based on their similarities with ideological reference texts. Our findings show that the RAG framework is prone to transferring ideological discourses into LLM responses, with sampling temperature having a measurable impact on the strength of this transfer. Discoursive alignment between generated answers and the reference text is highest at moderate temperatures, where models balance stochasticity with retrieval grounding, and drops at low temperatures, indicating that overly deterministic sampling suppresses discourse transfer.
Elmira Salari, H. Amamou, Jos'e Victor de Souza et al.· arXiv.org· 0 citations
Large Language Models (LLMs) can generate fluent and convincing responses, but fluency does not guarantee
factual correctness. Hallucination occurs when a model produces information that is false, unsupported, or inconsistent
with available evidence. This paper reviews why hallucinations arise andexamine Retrieval-Augmented Generation (RAG)
as a practical mitigation strategy. The discussion focuses on prediction-based generation, missing or outdated knowledge,
ambiguous queries, and the absence of automatic verification. RAG addresses these limitations by retrieving relevant
information from an external knowledge source and supplying it to the generator as contextual evidence. The paper
describes the major stages of a RAG pipeline, including document ingestion, chunking, embeddings, vector storage,
retrieval, context augmentation, and answer generation. It also examines the conditions under which RAG can fail,
including poor retrieval, incomplete knowledge bases, unreliable sources, weak ranking, and incorrect interpretation of
retrieved context. Finally, the paper discusses evaluation dimensions such as context relevance, answer faithfulness, and
answer relevance, and outlines applications of RAG in education, enterprise search, technical support, research assistance,
and document question answering. The analysis concludes that RAG should be viewed as a grounding and evidence-access
mechanism rather than a guarantee of hallucination-free generation
Shyalaja L. N., Shantinath Patil, Pruthviraj S. R. et al.· International Journal for Re...· 0 citations
HallDetect, a lightweight, reference-free, and black-box framework for hallucination detection, is presented, a lightweight, reference-free, and black-box framework for hallucination detection that is evaluated not only on summarization but across a broader range of source-grounded generation settings.
Achir Oukelmoun, N. Semmar, Gäel de Chalendar· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
