Adversarial training for named entity recognition of rail fault text

At present, most rail faults in metro systems are recorded in the form of text. Due to the lack of effective mining and analysis tools, information in the massive textual data is not fully utilized. Learning from past fault texts and identifying some key concepts are essential to analyze faults and help decision making. In this paper, a word-enhanced adversarial training model (AT-BiLSTM-CRF) is proposed to address this problem. In this model, the named entity recognition (NER) is achieved by bi-directional long short-term memory (BiLSTM) with conditional random field (CRF). At the same time, the Chinese word segmentation (CWS) task is introduced to conduct adversarial training with the NER task. The structure of adversarial training is to make full use of the boundary information and filter out the noise caused by introducing the CWS task. More importantly, the experiments on five different train fault datasets are conducted in the rail field. The results show that the model performs better than the state-of-the-art baselines, which indicates it has the potential to lay the foundation for textual data analysis in the rail field.