The Medicinal Herb–Disease Relationships (MHDR) Corpus

Traditional medicine (TM) has long utilized medicinal herbs (MHs) to prevent and treat diverse health conditions. In recent years, growing efforts to integrate TM with modern biomedical research have sought to validate traditional knowledge and uncover novel therapeutic insights. However, the exponential increase in biomedical publications has made it increasingly difficult to extract structured and interpretable knowledge—particularly relationships between MHs and diseases—from unstructured text. Although Natural Language Processing (NLP) and machine learning (ML) methods offer potential solutions, they remain limited by linguistic ambiguity, inconsistent terminology, and context-dependent meaning.<br>To address these challenges, we present the Medicinal Herb–Disease Relationships (MHDR) corpus, an expert-annotated resource designed to support computational analysis of MH–disease associations within the integrative TM–biomedicine domain. Derived from 800 PubMed abstracts, the MHDR corpus captures pharmacognostic-level herb mentions, reflecting plant parts and processing methods critical for precise entity normalization. The resulting dataset comprises 5,119 MH mentions, 6,621 disease mentions, and 1,314 annotated MH–disease relations extracted from 832 key sentences bearing evidences.<br>Baseline evaluations using Transformer-based language models demonstrate the corpus’s utility for relation extraction tasks in biomedical text mining. By providing a well-curated and semantically accurate dataset, the MHDR corpus contributes to advancing biomedical informatics research, enabling more accurate knowledge discovery, ontology integration, and computational modeling of therapeutic mechanisms in traditional medicine.