Semi-supervised sentence classification for MEDLINE documents(Medical Data Mining)

概要

論文の詳細を見る
We address the task of sentence classification in Medline abstracts, in which sentences must be classified into their structural roles such as background, objective, methods, experimental results, and conclusions. With a plenty of labeled data, supervised learning would be able to accurately infer the structural roles of each sentence in the abstracts. However, it is not practical to assume abundant training data as they are expensive to construct. We therefore apply semi-supervised learning to this sentence classification task to remedy the lack of training data. Experimental results show that semi-supervised learning outperform pure supervised learning, when only a small amount of correctly labeled sentences are available.
一般社団法人情報処理学会の論文
2004-12-04