A hybrid neural network model for predicting charges by incorporating legal knowledge

Abstract:
The charge prediction task is one of the hot research topics in legal intelligence. Its main objective is to assist judges in case decision-making by using the given case and then predicting the charges. Nowadays, introducing a deep learning framework to solve the legal intelligence topic is the mainstream approach. We propose data augmentation by incorporating knowledge of the law articles and combining it with the ERNIE_DPCNN hybrid model to enhance the charge prediction to the best effect through extensive research on legal documents. Moreover, we combine the Integrated Gradient to interpretability analysis of the prediction model to analyze the charge prediction results. In this paper, we use the accuracy rate, recall rate, and F1 value evaluation indexes for evaluation, and the experimental results show that the method can effectively predict the charges for legal cases.