A Model for Drug Discovery on Unstructured Text using Semi-Supervised Learning and Fuzzy Matching

Health related discoveries are mainly published as journal publications and the rate at which they are generated increases as new information and discoveries emerge. Discovery of latent medically-related terms in a document corpus is a challenging task where the researcher is not an expert in that domain and a viable database of medicine related words is not readily available. The objective of the study was to investigate the methodologies and best practices that enable discovery of latent drug terms found in health publications corpus for effective dissemination at county and national levels. Fuzzy matching methodology was considered for its near and exact matching algorithms. DrugBank dataset was chosen as reference for drug terms because of its comprehensive list of drugs, that are frequently updated and freely accessible. Semi-supervised learning was applied in modeling of multi-search medical terms on an hourly basis. Drug-name recognition, sentence categorization and information retrieval are among the features described in the presented model.