Legal judgment prediction is becoming a research hotspot in the legal field as an important artificial intelligence-assisted decision-making tool in legal case management, which is able to predict judgment results. In this paper, data from the 2018 China Law Research Cup competition is gathered, and the dataset is preprocessed in the context of international economic law. Then, a multi-task model for legal verdict prediction is proposed, and the training optimization and prediction of the model are designed using CNN, RNN, and LSTM as the semantic coding layer. The model proposed in this paper achieves a significant improvement of 8% and 6% in the accuracy of the model in the prediction of the charging task and the legal sentence task, respectively. In case outcome prediction, the accuracy of the model proposed in this paper is improved by 14.6% on average compared to the feature model-based modeling approach.