MedCorpus

MedCorpus is a large-scale benchmark dataset designed to advance research in personalized, multi-turn conversational literature retrieval within the biomedical domain. The dataset comprises 1,998 three-turn conversational sessions generated via a large language model to simulate realistic information-seeking behaviors. These dialogues are paired with a curated corpus of 28,500 high-impact biomedical articles from PubMed, and each query is annotated with silver-standard relevance labels, making MedCorpus a valuable resource for developing and evaluating next-generation conversational search systems.