Circumstance-Aware Graph Neural Network for Legal Judgment Prediction

Wenjing Pan, Yuan Chen, Zhiheng Liu, Xia Li, Zhuopeng Xu

Abstract: Legal judgment prediction aims to predict applicable law articles, charges, and the terms of penalty automatically based on the fact descriptions of the criminal cases. In general, law articles specify different penalties for the same charge according to different circumstances. When using law articles, it is important to determine the right circumstance corresponding to the fact, and apply the adequate penalty. However, most of the existing work does not distinguish circumstances in law articles, and uses the entire law article as a whole. In order to make finer-grained reasoning, we consider law articles at circumstance level. As descriptions of circumstances may depend on some other circumstances, we construct a circumstance-based law article graph (CLAG) reflecting the dependencies between circumstances in order to create better representations for them. We also propose a novel two-way attention mechanism between facts and circumstances, through which the representations of facts and circumstances can enhance mutually. The experimental results conducted on real world dataset show the superiority of our model.