Research on the Named Entity Recognition for Rail Fault Text Based on Distant Supervision

Most faults in rail field are recorded as texts, and neural network requires a large amount of labeled data which is used to mine and analyse the texts. However, manually labeled datasets are costly to obtain, so it is necessary to train a better model capable of recognising entities from small batches of manually annotated data. In this paper, a named entity recognition model based on large batches of distantly supervised data and small batches of manually annotated datasets is proposed, which increased the character representation. A reinforcement learning selector is used in the model to filter the distantly supervised data and a BERT encoder is implemented to enhance the character representation capability. Finally, the experiments on a real railway fault datasets are conducted with our proposed model, and the result shows that the model proposed in this paper outperforms other baseline models significantly, and is more adaptive with both reduced datasets.