Structural Hallucination in LLMs: A Formal Characterization and Mitigation Method

Author(s)	Mohan Siva Krishna Konakanchi
Country	United States
Abstract	Large Language Models (LLMs) have advanced natural language processing, yet they suffer from hallucinations—outputs that seem plausible but lack factual grounding. This paper identifies a novel category, ”structural hallucinations,” where generated content mimics the syntactic and semantic structure of truthful information without verifiable basis. We formally characterize these hallucinations using graph-theoretic semantic representations and probabilistic verifiability measures. To mitigate them, we propose a trust metric-based federated learning (FL) framework that ensures integrity and accountability across distributed data silos. This framework introduces a trust score balancing explainability and performance, optimized via multi-objective reinforcement learning. We quantify the trade-off between explainability and performance, enabling precise control over hallucination risks. Extensive experiments on benchmark datasets demonstrate a 45% reduction in structural hallucinations while maintaining high performance. This work enhances the reliability of LLMs, with implications for highstakes applications.
Keywords	Large Language Models, Structural Hallucination, Federated Learning, Trust Metrics, Explainability, Performance Trade-off
Field	Engineering
Published In	Volume 2, Issue 5, September-October 2020
Published On	2020-09-10
DOI	https://doi.org/10.36948/ijfmr.2020.v02i05.61072

About IJFMR Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Reviewer Referral Program Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us		+91-9687-828-838	editor@ijfmr.com

International Journal For Multidisciplinary Research