A Formal Approach to LLM Hallucination Detection and Mitigation
2026 (English)In: 2026 11th International Conference on Cloud Computing and Big Data Analytics (ICCCBDA), Institute of Electrical and Electronics Engineers Inc. , 2026, p. 678-685Conference paper, Published paper (Refereed)
Abstract [en]
This study focuses on 'LLM hallucinations'-a core risk in real-world multi-team projects-and provides its definition, classification, formalized description, as well as an initial approach to mitigation of LLM hallucinations. Through experiments, we demonstrate that hallucinations are an inherent deficit of language models which tend to 'over-complete' the users' requirements and can be effectively mitigated by using structured instructions combined with systematic reviews. This research provides usable theoretical frameworks and practical methodologies for improving the factual reliability of LLM-generated content in project-oriented scenarios. © 2026 IEEE.
Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers Inc. , 2026. p. 678-685
Keywords [en]
Formal Definition, Hallucination, LLMs, Mitigation Strategy, Formal approach, Formalized description, Language model, LLM, Over-complete, Real-world, Team projects, Formal methods
National Category
Computer Sciences
Identifiers
URN: urn:nbn:se:du-55036DOI: 10.1109/ICCCBDA69573.2026.11676510Scopus ID: 2-s2.0-105051520375ISBN: 9798331550264 (electronic)OAI: oai:DiVA.org:du-55036DiVA, id: diva2:2111913
Conference
11th International Conference on Cloud Computing and Big Data Analytics, ICCCBDA 2026, Chengdu, China, 24-26 April 2026
2026-10-212026-10-212026-10-21Bibliographically approved