[1N4-J-9] Natural language processing, information retrieval: domain knowledge analysis

[1N4-J-9-01] Chemical Named Entity Recognition with Self-Training

〇Yiming Cui^1,3, Hitoshi Nishikawa^1,3, Takenobu Tokunaga¹, Hiyori Yoshikawa^2,3, Tomoya Iwakura^2,3 (1. School of Computing, Tokyo Institute of Technology, 2. Fujitsu Laboratories Ltd., 3. RIKEN AIP-Fujitsu Collaboration Center)

Keywords:Named entity recognition, Neural network

In this paper, we propose to use self-training for chemical named entity recognition. We first train a neural network-based model for chemical named entity recognition model using the CHEMDNER corpus. The trained model is used to annotate the unlabelled MEDLINE corpus to create automatically labelled training data. We then use both training data, manually labelled CHEMNER corpus and automatically labelled MEDLINE corpus, to train our final model. The evaluation using the unlabelled MEDLINE corpus as test data showed that the effectiveness of self-training in the chemical named entity recognition task.