Element-Aware Legal Judgment Prediction for Criminal Cases with Confusing Charges

Abstract

Legal judgment prediction (LJP) plays an important role in legal assistant systems and aims to provide feasible judgment suggestions, including the charges, applicable law articles, and prison term. In practice, there exist many confusing charges which result in the decline of LJP performance of the existing works. To address this issue, we introduce the legal constitutive elements as the discriminative features to distinguish confusing charges. We propose an element-driven attentive neural network model, EDA-NN, which takes the textual description of a criminal case as the input and learns both element-free and element-aware case representations. Moreover, the element-driven attention mechanism is incorporated with the hierarchical sequence encoders, to generate crucial representations oriented to the legal constitutive elements at both the word and sentence levels. With the concatenation of element-free and element-aware representations, the EDA-NN can jointly predict the legal constitutive elements and judgment results. The experiments are conducted on a real-world dataset of criminal cases in mainland China. The experimental results demonstrate that our approach significantly outperforms all the baseline models on the LJP task for criminal cases with confusing cases.