Exploring the Usage of Knowledge Graphs in Identifying Human and LLM-Generated Fake Reviews

Ming Liu, Massimo Poesio


Abstract
The emergence of large language models has led to an explosion of machine-generated fake reviews. Although distinguishing between human and LLM-generated fake reviews is an area of active research, progress is still needed. One aspect which makes current LLM-generated fake reviews easier to recognize is that LLMs–in particular the smaller ones–lack domain-related knowledge. The objective of this work is to investigate whether large language models can produce more realistic artificial reviews when supplemented with knowledge graph information, thus resulting in a more challenging training dataset for human and LLM-generated fake reviews detectors. We propose a method for generating fake reviews by providing knowledge graph information to a llama model, and used it to generate a large number of fake reviews which used to fine tune a state-of-the-art human and LLM-generated fake reviews detection system. Our results show that when knowledge graph information is provided as part of the input, the accuracy of the model is improved by 0.24%. When the knowledge graph is used as an embedding layer and combined with the existing input embedding layer, the accuracy of the detection model is improved by 1.279%.
Anthology ID:
2025.ranlp-1.78
Volume:
Proceedings of the 15th International Conference on Recent Advances in Natural Language Processing - Natural Language Processing in the Generative AI Era
Month:
September
Year:
2025
Address:
Varna, Bulgaria
Editors:
Galia Angelova, Maria Kunilovskaya, Marie Escribe, Ruslan Mitkov
Venue:
RANLP
SIG:
Publisher:
INCOMA Ltd., Shoumen, Bulgaria
Note:
Pages:
674–681
Language:
URL:
https://aclanthology.org/2025.ranlp-1.78/
DOI:
Bibkey:
Cite (ACL):
Ming Liu and Massimo Poesio. 2025. Exploring the Usage of Knowledge Graphs in Identifying Human and LLM-Generated Fake Reviews. In Proceedings of the 15th International Conference on Recent Advances in Natural Language Processing - Natural Language Processing in the Generative AI Era, pages 674–681, Varna, Bulgaria. INCOMA Ltd., Shoumen, Bulgaria.
Cite (Informal):
Exploring the Usage of Knowledge Graphs in Identifying Human and LLM-Generated Fake Reviews (Liu & Poesio, RANLP 2025)
Copy Citation:
PDF:
https://aclanthology.org/2025.ranlp-1.78.pdf