Interpretable Charge Prediction with Multi-Perspective Jointly Learning Model
Xiaotong Yang, Guozhen Shi, Jiapeng Lou, Shubei Wang, Zichen Guo · 2019
Interpretable charge prediction refers to making a judgment interpretation with legal effects while predicting the charge according to the case, which can help people intuitively understand reasons of the judgment. Existing works only focus on the performance improvement of either charge prediction or judgment interpretation generation. However, the judgment interpretation that eliminates redundant information is conducive to charge prediction and the charge can guide the generation of the judgment interpretation. In addition, in order to make the charge-discriminative judgment interpretation less stereotyped, details in the case need to be paid attention to. Accordingly, we explore a dual-encoder to model the case from multi-perspective to extract richer information and establish two bridges between these two tasks to improve both of them so as to achieve interpretable charge prediction with a jointly learning model. Experimental results show that our model outperforms all strong baselines, improving the accuracy of charge prediction and generating flexible judgment interpretations.