Exploring Multi-Layer Convolutional Neural Networks for Railway Safety Text Classification

Abstract

With the rapid development of China High-Speed Rail, massive text data related to railway safety has been accumulated. When analyzing and understanding this data, classifying railway accident report text is essential and tedious work. Usually, such classification tasks are manually done by experts and workers in the railway safety department. Traditional data mining algorithms have been applied in these tasks to classify the text automatically. However, due to the complexity of the text data, classification algorithms sometimes fail and have insufficient learning ability. Meanwhile, the rise of machine learning enables us to deal with these complex problems effectively. In this paper, we propose an end-to-end multi-layer convolutional neural networks model to classify the railway safety-related text. We update the CNN part of the traditional model by increasing layers and adding a multi-height convolutional kernel. Additionally, we develop a data-preprocessing strategy to obtain the neat input data and reduce the complexity of the task. Experiments show that our proposed method achieves competitive performance and is suitable for railway safety-related text classification problems.