A Method of Extracting Sentences Containing Protein Function Information from Articles by Iterative Learning with Feature Update

摘要

Proteins are important macromolecules in living systems and serve various functions in almost all biological processes. Protein function information is reported in many scientific articles. Extraction of the function information from the articles is useful for drug discovery, understanding of life phenomenon, and so on. However, it is infeasible to extract the function information manually from a number of articles. In this paper, we propose a method of extracting sentences containing protein function information by iterative learning with feature update. In this method, we use a classifier in order to distinguish the sentences containing the function information from the other sentences, and introduce a semi-automatic procedure, in which a new classifier is reconstructed based on the user's feedback for the previous classified results. In the experiment with twelve articles as feedback data, it was confirmed that F-measure was improved by iterating learning without getting the negative effect of the feedback.