Graph Alignment Topology as an Inductive Bias for Grounding Detection

2026-05-21

AI Summary

Large language models (LLMs) often generate hallucinated content that contradicts factual information in reference documents, limiting their applicability in high-reliability settings. This work proposes a novel approach that constructs a bipartite alignment graph between reference texts and generated outputs and, for the first time, incorporates the topological structure of this alignment as an inductive bias for hallucination detection. By leveraging graph neural networks with message-passing mechanisms, the method learns structured alignment relationships without relying on retrieval augmentation or self-consistency heuristics commonly used in prior work. Evaluated across four diverse datasets for hallucination detection and question answering, the proposed approach achieves state-of-the-art performance, significantly outperforming mainstream LLMs including GPT-4o.

Abstract

Large Language Models (LLMs) are optimized to produce distributionally plausible continuations rather than to explicitly verify whether generated propositions are entailed by source documents. This inductive bias enables generalization, but it does not encode whether responses are grounded with respect to a reference. These issues limit the use of LLMs in domains where strict factual correctness is crucial, such as clinical decision support. Existing hallucination detection approaches improve factuality through retrieval augmentation, self-consistency, or claim verification, but generally do not learn directly over alignment topology. To leverage alignment topology as an inductive bias, we construct aligned bipartite graphs between reference information and LLM outputs and train a graph neural network (GNN) to model alignment structure using message passing. The method achieves state-of-the-art results on four diverse hallucination and question-answering datasets, outperforming all compared methods, including foundational LLMs such as GPT-4o.

Problem

• hallucination detection
• grounding
• large language models
• factual correctness
• alignment topology

Innovation

• graph alignment topology
• inductive bias
• grounding detection
• graph neural network
• hallucination detection