HART: Data-Driven Hallucination Attribution and Evidence-Based Tracing for Large Language Models

Shize Liang; Hongzhi Wang

cs.CLMar 6, 2026

HART: Data-Driven Hallucination Attribution and Evidence-Based Tracing for Large Language Models

Authors: Shize Liang, Hongzhi Wang

Organizations: Faculty of Computing, Harbin Institute of Technology, Harbin, China

Abstract

Large language models (LLMs) have demonstrated remarkable performance in text generation and knowledge-intensive question answering. Nevertheless, they are prone to producing hallucinated content, which severely undermines their reliability in high-stakes application domains. Existing hallucination attribution approaches, based on either external knowledge retrieval or internal model mechanisms, primarily focus on semantic similarity matching or representation-level discrimination. As a result, they have difficulty establishing structured correspondences at the span level between hallucination types, underlying error generation mechanisms, and external factual evidence, thereby limiting the interpretability of hallucinated fragments and the traceability of supporting or opposing evidence. To address these limitations, we propose HART, a fine-grained hallucination attribution and evidence retrieval framework for large language models. HART formalizes hallucination tracing as a structured modeling task comprising four stages: span localization, mechanism attribution, evidence retrieval, and causal tracing. Based upon this formulation, we develop the first structured dataset tailored for hallucination tracing, in which hallucination types, error mechanisms, and sets of counterfactual evidence are jointly annotated to enable causal-level interpretability evaluation. Experimental results on the proposed dataset demonstrate that HART substantially outperforms strong retrieval baselines, including BM25 and DPR, validating the effectiveness and generalization capability of the proposed tracing paradigm for hallucination analysis and evidence alignment.

Figures & tables

图 1. The construction of the hallucination tracing dataset employs a labeling framework combining large language model assistance with human supervision.

图 2. Evidence Retrieval Process: Leveraging semantic embeddings, vector-based indexing, and a Cross-Encoder fine-ranking mechanism, this process aligns hallucinated fragments with external factual evidence and retrieves the most relevant supporting information.

图 3. Span-level hallucination tracing workflow: HART integrates hallucination type classification, error mechanism attribution, and evidence retrieval to systematically model and trace the external evidence supporting each hallucinated segment in model-generated text.

Method	R@1	R@5	R@10	MRR	nDCG@5	nDCG@10
Dense Embedding	0.4133	0.6387	0.7146	0.5092	0.5340	0.5586
Dense Embedding + Cross-Encoder	0.5172	0.6868	0.7146	0.5887	0.6104	0.6196
Dense Embedding + Multi-Query	0.6244	0.8058	0.8557	0.7016	0.7227	0.7389
Ours (HART)	0.7068	0.8360	0.8557	0.7619	0.7786	0.7850

表 1. Ablation Study on Evidence Retrieval Performance

Category	SubType	Qwen( $\%$ )	Mistral(%)
Hallucination Type	Logic	0.0033	0.0087
	Fabricate	0.1028	0.0605
	Entity	0.1676	0.0999
	Fact	0.7263	0.8309
Error Mechanism	Context Leakage	0.0054	0.0076
	Reasoning Failure	0.0172	0.0168

表 2. Statistical Analysis of Hallucination Types and Error Mechanism Proportions in Qwen and Mistral Models

Panel of Figure 6 — 图 4. Loss and Accuracy During Classifier Training

Dataset	$k$	Method	Recall@ $k$	nDCG@ $k$	Joint SR@ $k$	MRR @ $k$
Qwen	$k=1$	BM25	0.1074	0.2659	0.0032	0.3016
		DPR	0.0349	0.0869	0.0014	0.1353
		Sentence-BERT	0.0859	0.2164	0.0020	0.1642
		Cross-Encoder	0.0906	0.2269	0.0025	0.3298
		Ours (HART)	0.8024	0.7915	0.6265	0.8024
	$k=2$	BM25	0.2196	0.3896	0.0043	0.3612

表 3. Hallucination Tracing Performance On Qwen And Mistral Dataset: HART vs. Baselines

Hallucination in Language Models

Hallucination in language models studies how systems produce unsupported or false claims and how to detect, explain, and reduce them.

See all 254 papers

Apr 6, 2026

HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders

Large Language Models (LLMs) are powerful and widely adopted, but their practical impact is limited by the well-known hallucination phenomenon. While recent hallucination detection methods have made notable progress, we find most of them overlook the dynamic nature and underlying mechanisms of it. To address this gap, we propose HalluSAE, a phase transition-inspired framework that models hallucination as a critical shift in the model's latent dynamics. By modeling the generation process as a trajectory through a potential energy landscape, HalluSAE identifies critical transition zones and attributes factual errors to specific high-energy sparse features. Our approach consists of three stages: (1) Potential Energy Empowered Phase Zone Localization via sparse autoencoders and a geometric potential energy metric; (2) Hallucination-related Sparse Feature Attribution using contrastive logit attribution; and (3) Probing-based Causal Hallucination Detection through linear probes on disentangled features. Extensive experiments on Gemma-2-9B demonstrate that HalluSAE achieves state-of-the-art hallucination detection performance.

Boshui Chen, Zhaoxin Fan, Ke Wang +5

Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing, School of Artificial Intelligence, Beihang University, Beijing, China. · State Key Laboratory of Virtual Reality Technology and Systems, Beihang University, Beijing, China. · Beijing Academy of Blockchain and Edge Computing, Beijing, China. +1

LLM Hallucination Detection

Ask PDF

Apr 13, 2026

CausalGaze: Unveiling Hallucinations via Counterfactual Graph Intervention in Large Language Models

Despite the groundbreaking advancements made by large language models (LLMs), hallucination remains a critical bottleneck for their deployment in high-stakes domains. Existing classification-based methods mainly rely on static and passive signals from internal states, which often captures the noise and spurious correlations, while overlooking the underlying causal mechanisms. To address this limitation, we shift the paradigm from passive observation to active intervention by introducing CausalGaze, a novel hallucination detection framework based on structural causal models (SCMs). CausalGaze models LLMs' internal states as dynamic causal graphs and employs counterfactual interventions to disentangle causal reasoning paths from incidental noise, thereby enhancing model interpretability. Extensive experiments across four datasets and three widely used LLMs demonstrate the effectiveness of CausalGaze, especially achieving 3.3% improvement in AUROC on the TruthfulQA dataset compared to state-of-the-art baselines.

Linggang Kong, Lei Wu, Yunlong Zhang +4

College of Electronic Engineering, National University of Defense Technology · Anhui Province Key Laboratory of Cyberspace Security Situation Awareness and Evaluation · Institute of Computer Application, China Academy of Engineering Physics

LLM Hallucination Detection Hallucination in Language Models Causal Interventions in Language Models LLM Interpretability

Ask PDF

Aug 16, 2026

Hallucination Span Detection with Input-Side Evidence Alignment

Hallucinations remain a major obstacle to the reliable use of large language models (LLMs) in conditional text generation. Existing methods primarily assess the factuality of an entire generated text, providing limited insight into which output spans are hallucinated or how they relate to the input. We introduce the task of hallucination span detection with input-side evidence alignment, which jointly identifies hallucinated spans and aligns output tokens with the corresponding input evidence. Our approach is based on the observation that faithful output tokens are predictable from the input, whereas hallucinated tokens are not. We therefore train an encoder-based model to predict masked output tokens from the input representation, using prediction confidence for hallucination detection while naturally producing alignments to the input. Experiments show that the proposed method effectively detects hallucinated spans and identifies meaningful input-side evidence. Human evaluation confirms the quality of the predicted alignments.

Miyu Yamada, Yuki Arase

Institute of Science Tokyo

Hallucination Detection LLM Hallucination Detection

Ask PDF

Abstract

Hallucination in Language Models

Explore similar work

HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders

CausalGaze: Unveiling Hallucinations via Counterfactual Graph Intervention in Large Language Models

Hallucination Span Detection with Input-Side Evidence Alignment