This paper presents a retrieval-based hallucination detection system for Large Language Models (LLMs). The system verifies generated responses using Wikipedia-based knowledge and semantic similarity techniques. Experimental evaluation on a dataset of 100 questions achieved 98% accuracy with high recall and an F1-score of 0.75. The system also demonstrates performance variations in real-world scenarios due to retrieval limitations and semantic ambiguity. The proposed approach highlights the importance of integrating external knowledge verification with generative AI systems to improve reliability and factual correctness.
Bhatt et al. (Wed,) studied this question.