Case Retrieval for Legal Judgment Prediction in Legal Artificial Intelligence

Case Retrieval for Legal Judgment Prediction in Legal Artificial Intelligence Han Zhang School of Information, Renmin University of China. zhanghanjl@ruc.edu.cn Zhicheng Dou Gaoling School of Artificial Intelligence, Renmin University of China. dou@ruc.edu.cn Abstract Legal judgment prediction (LJP) is a basic task in legal artificial intelligence. It consists of three subtasks, which are relevant law article prediction, charge prediction and term of penalty prediction, and gives the judgment results to assist the work of judges. In recent years, many deep learning methods have emerged to improve the performance of the legal judgment prediction task. The previous methods mainly improve the performance by integrating law articles and the fact description of a legal case. However, they rarely consider that the judges usually look up historical cases before making a judgment in the actual scenario. To simulate this scenario, we propose a historical case retrieval framework for the legal judgment prediction task. Specifically, we select some historical cases which include all categories from the training dataset. Then, we retrieve the most similar Top-k historical cases of the current legal case and use the vector representation of these Top-k historical cases to help predict the judgment results. On two real-world legal datasets, our model achieves better results than several state-of-the-art baseline models. 1 Introduction With the rapid development of artificial intelligence, it has become a trend to use artificial intelligence to help judicial personnel. Legal judgment prediction (LJP) is such an artificial intelligence task in legal artificial intelligence. As shown in Table 1, given the fact description of a legal case, the legal judgment prediction task can provide the judgment result of the case. The predicted result consists of three parts: relevant law article, charge and term of penalty. Legal judgment prediction can not only give the judgment results efficiently for reference for judicial personnel but also provide legal suggestions for ordinary people when there is a legal dispute (Zhong et al., 2020; Wang et al., 2019; Zhong et al., 2018; Zhang et al., 2021) in daily life. With the application of deep learning in legal artificial intelligence, various methods have been proposed to improve the performance of legal judgment prediction. Some methods (Zhong et al., 2018; Yang et al., 2019) consider using the order information among the three subtasks of legal judgment prediction in reality to improve the representation of fact description. Further, some methods (Yue et al., 2021; Ma et al., 2021; Feng et al., 2022) consider a fine-grained division of the fact description to improve the fact representation. Additionally, some methods (Luo et al., 2017; Hu et al., 2018; Wang et al., 2019; Xu et al., 2020) consider the important role of law articles in reality and introduce them to improve the performance. These efforts have effectively improved the performance of legal judgment prediction. However, the existing methods are affected by the fact that the law articles are too concise and still have limitations in modelling the judgment process. On the one hand, the law articles are very concise and lack …

Content truncated to 500 words