We focus on named entity recognition (NER) for Chinese social media. Withmassive unlabeled text and quite limited labelled corpus, we propose asemi-supervised learning model based on B-LSTM neural network. To takeadvantage of traditional methods in NER such as CRF, we combine transitionprobability with deep learning in our model. To bridge the gap between labelaccuracy and F-score of NER, we construct a model which can be directly trainedon F-score. When considering the instability of F-score driven method andmeaningful information provided by label accuracy, we propose an integratedmethod to train on both F-score and label accuracy. Our integrated model yields7.44\% improvement over previous state-of-the-art result.
展开▼
