Ontology Enrichment from Free-text Clinical Documents: A Comparison of Alternative Approaches.

摘要

While the biomedical informatics community widely acknowledges the utility of domain ontologies, there remain many barriers to their effective use. One important requirement of domain ontologies is that they achieve a high degree of coverage of the domain concepts and concept relationships. However, the development of these ontologies is typically a manual, time-consuming, and often error-prone process. Limited resources result in missing concepts and relationships, as well as difficulty in updating the ontology as domain knowledge changes. Methodologies developed in the fields of Natural Language Processing (NLP), Information Extraction (IE), Information Retrieval (IR), and Machine Learning (ML) provide techniques for automating the enrichment of ontology from free-text documents. In this dissertation, I extended these methodologies into biomedical ontology development. First, I reviewed existing methodologies and systems developed in the fields of NLP, IR, and IE, and discussed how existing methods can benefit the development of biomedical ontologies. This previously unconducted review was published in the Journal of Biomedical Informatics. Second, I compared the effectiveness of three methods from two different approaches, the symbolic (the Hearst method) and the statistical (the Church and Lin methods), using clinical free-text documents. Third, I developed a methodological framework for Ontology Learning (OL) evaluation and comparison. This framework permits evaluation of the two types of OL approaches that include three OL methods. The significance of this work is as follows: 1) The results from the comparative study showed the potential of these methods for biomedical ontology enrichment. For the two targeted domains (NCIT and RadLex), the Hearst method revealed an average of 21% and 11% new concept acceptance rates, respectively. The Lin method produced a 74% acceptance rate for NCIT; the Church method, 53%. As a result of this study (published in the Journal of Methods of Information in Medicine), many suggested candidates have been incorporated into the NCIT; 2) The evaluation framework is flexible and general enough that it can analyze the performance of ontology enrichment methods for many domains, thus expediting the process of automation and minimizing the likelihood that key concepts and relationships would be missed as domain knowledge evolves.