Legal Judgment Prediction with Label Dependencies

Long Chen, Nuo Xu, Yue Wang

Abstract: Legal Judgment Prediction (LJP) is a key technique for social fair. It aims to predict the judicial decisions automatically given the fact description and has great prospects in judicial assistance and management. This article focuses on the prediction of criminal judgment and proposes a legal domain-oriented method for the LJP task, by exploiting the dependencies of labels across tasks of LJP. The proposed method captures the dependencies by a prediction forward-propagate mechanism over a directed heterogeneous graph, and a novel prediction task, attribute prediction. The experiments prove the efficiency of the method and show the superior of our model on real-world datasets.