Neural Message-Passing on Attention Graphs for Hallucination Detection

• URL: http://arxiv.org/abs/2509.24770v1
• Date: Mon, 29 Sep 2025 13:37:12 GMT
• Title: Neural Message-Passing on Attention Graphs for Hallucination Detection
• Authors: Fabrizio Frasca, Guy Bar-Shalom, Yftah Ziser, Haggai Maron,
• Abstract summary: CHARM casts hallucination detection as a graph learning task and tackles it by applying GNNs over the above attributed graphs.<n>We show that CHARM provably subsumes prior attention-based traces and, experimentally, it consistently outperforms other approaches across diverse benchmarks.
• Score: 32.29963721910821
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large Language Models (LLMs) often generate incorrect or unsupported content, known as hallucinations. Existing detection methods rely on heuristics or simple models over isolated computational traces such as activations, or attention maps. We unify these signals by representing them as attributed graphs, where tokens are nodes, edges follow attentional flows, and both carry features from attention scores and activations. Our approach, CHARM, casts hallucination detection as a graph learning task and tackles it by applying GNNs over the above attributed graphs. We show that CHARM provably subsumes prior attention-based heuristics and, experimentally, it consistently outperforms other leading approaches across diverse benchmarks. Our results shed light on the relevant role played by the graph structure and on the benefits of combining computational traces, whilst showing CHARM exhibits promising zero-shot performance on cross-dataset transfer.

Related papers

• Detecting Contextual Hallucinations in LLMs with Frequency-Aware Attention [27.49425252327799]
  We introduce a frequency-aware perspective on attention by analyzing its variation during generation.<n>We develop a lightweight hallucination detector using high-frequency attention features.
  arXiv  Detail & Related papers  (2026-02-20T11:18:45Z)
• TDGNet: Hallucination Detection in Diffusion Language Models via Temporal Dynamic Graphs [30.313604786976715]
  Diffusion language models (D-LLMs) offer parallel denoising and bidirectional context.<n> hallucination detection for D-LLMs remains underexplored.<n>We introduce TDGNet, a temporal dynamic graph framework that formulates hallucination detection as learning over evolving token-level attention graphs.
  arXiv  Detail & Related papers  (2026-02-08T16:35:30Z)
• Hallucination Detection in LLMs with Topological Divergence on Attention Graphs [64.74977204942199]
  Hallucination, i.e., generating factually incorrect content, remains a critical challenge for large language models.<n>We introduce TOHA, a TOpology-based HAllucination detector in the RAG setting.
  arXiv  Detail & Related papers  (2025-04-14T10:06:27Z)
• Hallucination Detection in LLMs Using Spectral Features of Attention Maps [8.820670807424174]
  Large Language Models (LLMs) have demonstrated remarkable performance across various tasks but remain prone to hallucinations.<n>Recent methods leverage attention map properties to this end, though their effectiveness remains limited.<n>We propose the $textLapEigvals$ method, which uses the top-$k$ eigenvalues of the Laplacian matrix derived from the attention maps as an input to hallucination detection probes.
  arXiv  Detail & Related papers  (2025-02-24T19:30:24Z)
• Beyond Message Passing: Neural Graph Pattern Machine [50.78679002846741]
  We introduce the Neural Graph Pattern Machine (GPM), a novel framework that bypasses message passing by learning directly from graph substructures.<n>GPM efficiently extracts, encodes, and prioritizes task-relevant graph patterns, offering greater expressivity and improved ability to capture long-range dependencies.
  arXiv  Detail & Related papers  (2025-01-30T20:37:47Z)
• Integrating a Heterogeneous Graph with Entity-aware Self-attention using Relative Position Labels for Reading Comprehension Model [14.721615285883429]
  We introduce a novel attention pattern that integrates reasoning knowledge derived from a heterogeneous graph into the transformer architecture without relying on external knowledge. The proposed attention pattern comprises three key elements: global-local attention for word tokens, graph attention for entity tokens that exhibit strong attention towards tokens connected in the graph, and the consideration of the type of relationship between each entity token and word token. Our model outperforms both the cutting-edge LUKE-Graph and the baseline LUKE model across two distinct datasets.
  arXiv  Detail & Related papers  (2023-07-19T20:17:37Z)
• Causally-guided Regularization of Graph Attention Improves Generalizability [69.09877209676266]
  We introduce CAR, a general-purpose regularization framework for graph attention networks. Methodname aligns the attention mechanism with the causal effects of active interventions on graph connectivity. For social media network-sized graphs, a CAR-guided graph rewiring approach could allow us to combine the scalability of graph convolutional methods with the higher performance of graph attention.
  arXiv  Detail & Related papers  (2022-10-20T01:29:10Z)
• Beyond the Gates of Euclidean Space: Temporal-Discrimination-Fusions and Attention-based Graph Neural Network for Human Activity Recognition [5.600003119721707]
  Human activity recognition (HAR) through wearable devices has received much interest due to its numerous applications in fitness tracking, wellness screening, and supported living. Traditional deep learning (DL) has set a state of the art performance for HAR domain. We propose an approach based on Graph Neural Networks (GNNs) for structuring the input representation and exploiting the relations among the samples.
  arXiv  Detail & Related papers  (2022-06-10T03:04:23Z)
• How to Find Your Friendly Neighborhood: Graph Attention Design with Self-Supervision [16.86132592140062]
  We propose a self-supervised graph attention network (SuperGAT) for noisy graphs. We exploit two attention forms compatible with a self-supervised task to predict edges. By encoding edges, SuperGAT learns more expressive attention in distinguishing mislinked neighbors.
  arXiv  Detail & Related papers  (2022-04-11T05:45:09Z)
• FDGATII : Fast Dynamic Graph Attention with Initial Residual and Identity Mapping [0.39373541926236766]
  We propose a novel graph neural network FDGATII, inspired by attention mechanism's ability to focus on selective information. By using sparse dynamic attention, FDG ATII is inherently parallelizable in design, whist efficient in operation. We show that FDG ATII outperforms GAT and GCN based benchmarks in accuracy and performance on fully supervised tasks.
  arXiv  Detail & Related papers  (2021-10-21T20:19:17Z)
• Joint Graph Learning and Matching for Semantic Feature Correspondence [69.71998282148762]
  We propose a joint emphgraph learning and matching network, named GLAM, to explore reliable graph structures for boosting graph matching. The proposed method is evaluated on three popular visual matching benchmarks (Pascal VOC, Willow Object and SPair-71k) It outperforms previous state-of-the-art graph matching methods by significant margins on all benchmarks.
  arXiv  Detail & Related papers  (2021-09-01T08:24:02Z)
• Multi-hop Attention Graph Neural Network [70.21119504298078]
  Multi-hop Attention Graph Neural Network (MAGNA) is a principled way to incorporate multi-hop context information into every layer of attention computation. We show that MAGNA captures large-scale structural information in every layer, and has a low-pass effect that eliminates noisy high-frequency information from graph data.
  arXiv  Detail & Related papers  (2020-09-29T22:41:19Z)
• A Graph-based Interactive Reasoning for Human-Object Interaction Detection [71.50535113279551]
  We present a novel graph-based interactive reasoning model called Interactive Graph (abbr. in-Graph) to infer HOIs. We construct a new framework to assemble in-Graph models for detecting HOIs, namely in-GraphNet. Our framework is end-to-end trainable and free from costly annotations like human pose.
  arXiv  Detail & Related papers  (2020-07-14T09:29:03Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.