SnoMedTagger: A Semantic Tagger for Medical Narratives

The identification and classification of semantic information in medical narratives is critical for various research applications such as question-answering systems, statistical analysis, etc. Our contribution is a novel semantic tagger named SnoMedTagger to tag complex semantic information (paraphrases of concepts, abbreviations of concepts, complex multiword concepts) with 16 SNOMED CT semantic categories in medical narratives. SnoMedTagger is developed to support domain users as well as non-domain users working on research questions using medical narratives. Our method includes corpus-based rule-patterns from real world dataset and rule-patterns developed by refinement of SNOMED CT (Systemised NOmenclature of MEDicine-Clinical Terms) clinical vocabulary. These rulepatterns were able to identify semantic information in a range of text and classify them with respective semantic categories derived from SNOMED CT. On unseen gold standard, our rulepattern-based semantic tagger outperformed SVM-based machine learning system and Ontology-based Bioportal web annotator. The study has shown that it is possible to identify and classify complete semantic information with SNOMED CT semantic categories in medical narratives with high accuracy than achieved by existing approaches. 82 SAMAN HINA, ERIC ATWELL, OWEN JOHNSON