Semantic Relation Discovery by Using Co-occurrence Information
Stefan Schulz, Catalina Martínez Costa, Markus Kreuzthaler, José Antonio Miñarro-Giménez, Ulrich Andersen, Anders Boeck Jensen, Bente Mægaard · 2014
Motivated by the need of constructing a knowledge base for a patient-centred question-answering system, the potential of exploiting co-occurrence data to infer non-ontological semantic relations out of these statistical associations is explored. The UMLS concept co-occurrence table MRCOC is used as a data source. This data provides, for each co-occurrence record, a profile of MeSH subheading profiles. This is used as an additional source of semantic information from which we generate hypotheses for more specific semantic relations. An initial experiment was performed, limited to the study of disease-substance associations. For validation 20 diseases were selected and annotated by experts regarding treatment and prevention. The results showed good precision values (82 for prevention, 72 for treatment), but unsatisfactory values for recall (67 for prevention, 45 for treatment) for this particular use case.