Semi-supervised biomedical relation classification using generalized expectation criteria

Semi-supervised learning is necessary for the extensive application of machine learning in practice. It can use unlabeled data to improve the performance of existing supervised machine learning method. In this work, we addressed the biomedical relation classification problem by utilizing a semi-supervised method which can train Maximum Entropy models according to Generalized Expectation criteria. In the proposed method, instead of "instance labeling" used in previous works, the "feature labeling" was applied to get the training data which can save lots of labeling time. A topic model was involved to choose the features for labeling. Experiment results show that the proposed method can dramatically improve the performance of biomedical relation classification through incorporating unlabeled data by feature labeling.