Combining supervised learning techniques to key-phrase extraction for biomedical full-text

Key-phrase extraction plays a useful a role in research areas of Information Systems (IS) like digital libraries. Short metadata like key phrases are beneficial for searchers to understand the concepts found in the documents. This paper evaluates the effectiveness of different supervised learning techniques on biomedical full-text: Sequential Minimal Optimization (SMO) and K-Nearest Neighbor, both of which could be embedded inside an information system for document search. The authors use these techniques to extract key phrases from PubMed and evaluate the performance of these systems using the holdout validation method. This paper compares different classifier techniques and performance differences between the full-text and it's abstract. Compared with the authors' previous work, which investigated the performance of Naïve Bayes, Linear Regression and SVM(reg1/2), this paper finds that SVMreg-1 performs best in key-phrase extraction for full-text, whereas Naïve Bayes performs best for abstracts. These techniques should be considered for use in information system search functionality. Additional research issues also are identified.