查看文章

In the digital age, many countries have digitalized their judgement documents and uploaded them online. The automated classification of these documents could save lots of manpower and time. In this paper, an automatic classification method is developed for online judgement documents from Washington University School of Law Supreme Court Database (SCDB). Our approach was designed under a hybrid framework of deep learning, which couples convolution neural network (CNN) and with a recurrent neural network called bidirectional long short-term memory (BiLSTM). Firstly, the CNN architecture was optimized by genetic algorithm to generate the optimal word vector. Then, the vector was used to train the bi-LSTM for Softmax classification. The experimental results show that our model far exceeded the existing models in classification accuracy, indicating the effectiveness of integrating the genetically …