Study on the Multi-Task Model for Legal Judgment Prediction

In this study, the multi-task learning(MTL) classification method based on CNN-BiGRU model is proposed, which is used to improve the accuracy and efficiently of legal judgment prediction. The subtasks of legal judgment prediction are law artivles, charges and the terms of penalty. However, the single task learning(STL) models are used to analyze legal documents, which ignoring the correlation among the subtasks. The MTL model of CNN-BiGRU enhance the task learning process, which can extract the shared information among subtasks and learn multiple tasks at the same time. Therefore, in view of the shorcomings of STL, this study explored the affilication of the MTL method to predict the three subtasks of legal judgment. CNN-BiGRU has combined the good extraction ability of CNN for local feature information and RNN for longterm dependencies information of the text classification. Compared with the CAIL2018-Small dataset, the accuracy and F1-score are the highest of all baselines models. The accuracy and F1-score of the law articles, charges and the terms of penalty are 95.1%,95.2%,72.6% and 95.2%, 95.4%, 72.7%, respectively. The proposed model improves the interpretability and the gneralization ability. The effectiveness and suitability of the model are validated on legal judgment prediction tasks.