Semantic Representations of Terms in Traditional Chinese Medicine

Abstract

Word embeddings have been widely used in lexical semantics and neural networks in Natural Language Processing. This article investigates the semantic representations using word embedding technologies by verifying them on a human constructed domain ontology. The domain of Traditional Chinese Medicine (TCM) is used as a workbench in this study, because this domain is knowledge-rich and has a large-scale domain ontology with well-defined entity types and relation types. This article releases a dataset, named “TCMSem”, to capture TCM domain experts’ intuitions of semantic relatedness. This data set is designed to cover the medical entities and relations with as many semantic types as possible so as to initiate a diverse and comprehensive evaluation on word embeddings. Experimental results show that word embeddings have demonstrated higher proficiencies in the detection of synonyms and collocations than other types of semantic relations. Furthermore, the semantic relatedness of thousands of terms of major categories in TCM is visualized using the taxonomy defined in the ontology.