Similar Case Based Prison Term Prediction.

Legal judgment (e.g., charge, law article and prison term) prediction is an important task for Legal AI, aiming to assist the judges to get the legal judgment and improve the efficiency of judges. Recently, many existing works have been proposed to promote the performance of legal judgment prediction. While most of them pay attention to confusing charges identification and article recommendation, neglecting the prison term prediction could limit the overall performance (e.g., the accuracy of prison term is less than 50%). In this paper, we focus on the task of prison term prediction. According to the principle of "treating like cases alike" stressed by the Supreme People's Court of China, we propose a similar case based prison term prediction (SPTP) method, consisting of a prison term prediction module and a prison term rectification module. The prison term prediction module uses the basic prison term prediction method to get an initial prison term, and then the prison term rectification module rectifies the initial prison term to acquire the final prison term prediction. Specifically, the rectification module includes a similar case retrieval part and a sentencing rectification part. Extensive experiments show the effectiveness of our method under the quantitative evaluation metrics.