IEEE InCODE-2026In Reviewconference

Semantic Consistency for Hallucination Detection in Large Language Models

Charan Sai Ponnada, Dr. A. Reddy

IEEE International Conference on Computing, Communication and Intelligent Systems (InCODE) · April 1, 2026

Detection F1

0.89

Models Evaluated

5 LLMs

Benchmarks

3 Datasets

Improvement

+12% F1

Abstract

We propose a novel framework for detecting hallucinations in large language model outputs using semantic consistency analysis. Our approach generates multiple semantically equivalent paraphrases of model outputs and measures consistency across them using sentence-level embeddings and cross-attention mechanisms. Experimental results on multiple LLM benchmarks demonstrate that our method outperforms existing factuality metrics by a significant margin.

Keywords

Hallucination DetectionLarge Language ModelsSemantic ConsistencyNatural Language ProcessingAI Safety

Citation

@inproceedings{ponnada2026semantic,
  title={Semantic Consistency for Hallucination Detection in Large Language Models},
  author={Ponnada, Charan Sai and Reddy, A.},
  booktitle={Proceedings of IEEE InCODE 2026},
  year={2026}
}