Semantic grounding of LLMs using knowledge graphs for query reformulation in medical information retrieval

Abstract

Electronic health records contain abundant unstructured text, but long and noisy patient context can hinder effective document retrieval. This work evaluates a retrieval-augmented generation approach that integrates medical knowledge graphs with large language models for query reformulation. Experiments on two TREC benchmark datasets indicate that knowledge-graph grounding can improve the reliability and domain relevance of medical query refinement.

Publication
_2024 IEEE International Conference on Big Data (BigData), Washington, DC, USA, pp. 4048-4057, https://doi.org/10.1109/BigData62323.2024.10826117_