Domain Category Information as a Guide for Sentence Ranking to Support Medical Text Summarization

Medical professionals are required to pursue evidence-based practice by including the best available evidence from published research in their decision-making process. However, the exponential growth of biomedical resources makes it difficult for them to follow this requirement, and as a result, there is a need for automatic systems that can generate short, evidence-based summaries to assist them in the process. This thesis contributes to existing research by proposing sentence ranking strategies to incorporate domain-specific categorial information to support medical text summarization. Particularly, both structure-related categories (e.g., Population, Intervention, Background, other, Study, Outcome) and evidence-based categories (e.g., Strength of Recommendation Taxonomy) are considered. For the evidence-based categories, we developed our own SVM classifier so that we can compute the categorial information for all sentences in our dataset. These contributions are evaluated experimentally against an existing baseline system, along with a description about their potential applications.