Reducing Hallucinations in LLM-based Scientific Literature Analysis Using Peer Context Outlier Detection

Daniel Xie; Maxwell J. Jacobson; Adil Wazeer; Haiyan Wang; Xinghang Zhang; Yexiang Xue

cs.AIApr 1, 2026

Reducing Hallucinations in LLM-based Scientific Literature Analysis Using Peer Context Outlier Detection

Authors: Daniel Xie, Maxwell J. Jacobson, Adil Wazeer, Haiyan Wang, Xinghang Zhang, Yexiang Xue

Organizations: Department of Computer Science, Purdue University, IN, USA · School of Materials Engineering, Purdue University, IN, USA

Abstract

Reducing hallucinations in Large Language Models (LLMs) is essential for accurate data extraction from large text corpora. Current methods, like prompt engineering and chain-of-thought prompting, focus on individual documents and fail to consider relationships across a corpus. This paper introduces Peer Context Outlier Detection (P-COD), which uses inter-document relationships to improve extraction accuracy in scientific literature summarization, where papers with similar experiment settings should draw similar conclusions. By comparing extracted data to validated peer information within the corpus, we adjust confidence scores and flag low-confidence results for expert review. Our experiments demonstrate up to 98% precision in outlier detection across 6 scientific domains, reducing hallucinations and letting researchers focus on genuinely ambiguous cases.

Figures & tables

Fig. 1: High-level idea behind Peer Context Outlier Detection. Each point is a scientific paper; color indicates an experimental result value. Semantically similar papers cluster together. The black flag marks an outlier whose extracted value deviates from its peers, suggesting a likely LLM hallucination to flag for human review.

Fig. 2: Workflow of P-COD. An extraction LLM pulls relevant data from papers; a text-embedding LLM places papers in a shared semantic space where proximity reflects methodological similarity. Within each peer group, extracted values are compared and a surprising score is assigned based on deviation. Papers whose results diverge sharply from close peers are flagged for human review; color deviation makes outliers immediately visible.

Fig. 3: Multi-domain validation across six scientific fields. Each subplot shows semantic clustering and outlier detection with domain-specific metrics: CS (accuracy, 70–95%), Physics (wavelength, 400–800 nm), Biology (protein concentration, 2–1000 ng/mL), Chemistry (yield, 50–100%), Materials (tensile strength, 100–2000 MPa), and Environmental Science (CO 2 , 380–450 ppm). Points are individual papers positioned by semantic similarity, colored by experimental value. Black flags mark synthetic corrupted papers that P-COD identifies as peer outliers.

Fig. 4: Large-scale single-domain clustering across 8 CS sub-fields (200 papers, shared accuracy metric 0.70–0.95). Cluster separation is increased (factor 3.0) for visualization; black flags mark a representative sample of papers P-COD identified as outliers. Tight clustering shows CS papers are semantically closer than cross-domain papers, yet P-COD still precisely detects experimental anomalies within them.

	Precision (%)
Method	CS	Physics	Biology	Chemistry	Materials	Env. Sci.
P-COD (Ours)	98.0	66.7	80.0	100.0	87.5	100.0
Prompt Engineering	52.3	48.1	63.8	57.2	45.7	68.4
Chain-of-Thought	61.7	58.3	74.2	67.9	62.4	76.8

TABLE I: Precision of outlier detection methods across six scientific domains. CS is from the large-scale single-domain experiment.

LLM Hallucination Mitigation

Large language model hallucination mitigation develops ways to detect and reduce unsupported, fabricated, or inconsistent outputs.

See all 178 papers

MeasHalu: Mitigation of Scientific Measurement Hallucinations for Large Language Models with Enhanced Reasoning
Apr 18, 2026Ruijun Huang, Zhiqiao Kang, Yuxuan Zhu +5LLM Hallucination Mitigation Document Information Extraction
1Shenzhen Key Laboratory for High Performance Data Mining, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences · Artificial Intelligence Research Institute, Shenzhen University of Advanced Technology
HalluPeer: A Taxonomy-driven Benchmark for Detecting Hallucinations in Scientific Peer Reviews
Sep 3, 2026Tzu-Ling Lin, Dong-Ting Yao, Teng-Fang Hsiao +2Hallucination Detection Scholarly Peer Review
National Yang Ming Chiao Tung University
Self-prompting and cross-model consensus enable reproducible data extraction from scientific literature with large language models
Aug 19, 2026Valentin Romanov, Monique Bax, Steven NiedererLLM Information Extraction LLM Reliability
National Heart & Lung Institute, Faculty of Medicine, Imperial College London, United Kingdom · Cambridge Stem Cell Institute, JeDrey Cheah Biomedical Centre, Cambridge Biomedical Campus, University of Cambridge, Cambridge, United Kingdom · Department of Medicine, University of Cambridge, Cambridge, United Kingdom +1

Apr 18, 2026J/K move · Enter open · S save

MeasHalu: Mitigation of Scientific Measurement Hallucinations for Large Language Models with Enhanced Reasoning

Ruijun Huang, Zhiqiao Kang, Yuxuan Zhu, Junxiong Li+4

1Shenzhen Key Laboratory for High Performance Data Mining, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences · Artificial Intelligence Research Institute, Shenzhen University of Advanced Technology

LLM Hallucination Mitigation Document Information Extraction LLM Information Extraction

Open paper AskPDF arXiv

The accurate extraction of scientific measurements from literature is a critical yet challenging task in AI4Science, enabling large-scale analysis and integration of quantitative research findings. However, Large Language Models (LLMs) frequently exhibit severe hallucinations, which significantly undermine the reliability of automated scientific document understanding systems. To address this problem, we propose MeasHalu, a novel framework for mitigating scientific measurement hallucinations through enhanced reasoning and targeted optimization. We first present a fine-grained taxonomy of measurement-specific hallucinations, categorizing errors across quantities, units, modifiers, and relations. Our approach incorporates a two-stage reasoning-aware fine-tuning strategy using augmented scientific data and process-based supervision. Furthermore, we introduce a progressive reward curriculum designed to penalize specific hallucination types, significantly improving extraction faithfulness. Experimental results demonstrate that MeasHalu substantially reduces hallucination rates and improves overall accuracy on the MeasEval benchmark. This work provides a targeted solution to a key bottleneck in automated scientific knowledge extraction, facilitating more trustworthy and scalable machine-assisted scientific literature analysis.

Abstract

LLM Hallucination Mitigation

Explore similar work

MeasHalu: Mitigation of Scientific Measurement Hallucinations for Large Language Models with Enhanced Reasoning

HalluPeer: A Taxonomy-driven Benchmark for Detecting Hallucinations in Scientific Peer Reviews

Self-prompting and cross-model consensus enable reproducible data extraction from scientific literature with large language models