Biomedical concept extraction using concept graphs and ontology-based mapping

Assigning keywords to articles can be extremely costly. In this paper we propose a new approach to biomedical concept extraction using semantic features of concept graphs to help in automatic labeling of scientific publications. The proposed system extracts key concepts similar to author-provided keywords. We represent full-text documents by graphs and map biomedical terms to predefined ontology concepts. In addition to occurrence frequency weights, we use concept relation weights to rank potential key concepts. We compare our technique to that of KEA's, a state-of-the-art keyphrase extraction software. The results show that using the relations weight significantly improves the performance of concept extraction. The results also highlight the subjectivity of the concept extraction procedure as well as of its evaluation.