Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (3): 253-263.DOI: 10.3778/j.issn.1002-8331.2108-0340

• Engineering and Applications • Previous Articles     Next Articles

Integrating Multi-Layer Structure Information of Law for Legal Judgement Prediction

ZHANG Han, ZHENG Weihao, DOU Zhicheng, WEN Jirong   

  1. 1.School of Information, Renmin University of China, Beijing 100872, China
    2.Beijing Key Laboratory of Big Data Management and Analysis Methods, Beijing 100872, China
    3.Gaoling School of Artificial Intelligence, Renmin University of China, Beijing 100872, China
    4.Key Laboratory of Data Engineering and Knowledge Engineering, Beijing 100872, China
  • Online:2023-02-01 Published:2023-02-01

融合法律文本结构信息的刑事案件判决预测

张晗,郑伟昊,窦志成,文继荣   

  1. 1.中国人民大学 信息学院,北京 100872
    2.大数据管理与分析方法研究北京市重点实验室,北京 100872
    3.中国人民大学 高瓴人工智能学院,北京 100872
    4.数据工程与知识工程教育部重点实验室,北京 100872

Abstract: In recent years, the intellectualization of the judicial field has attracted extensive attention of academic circles. This paper chooses legal judgment prediction which is an important task in the judicial field as this research topic. And the legal judgment prediction includes three sub-tasks:law article recommendation, charge prediction and term of penalty prediction. With the wide application of deep learning in various fields, some researchers have introduced the deep learning method into the task of legal judgment prediction and have achieved good results. The existing methods of legal judgment prediction based on deep learning usually improve the model prediction ability by building the attention between case description and law articles or improve the overall performance by using the relationship of the sub-tasks. But these works do not consider the multi-layer structured information in the law. To solve the problem, this paper introduces the multi-layer structured information of legal texts to the task of legal judgment prediction. Specifically, this paper preprocesses the multi-layer structured information of legal texts and integrates the information of the law into the case description by using the co-attention mechanism and obtains the representation of the case description of each sub-task which integrates different levels of law information, so as to improve the performance of legal judgment prediction. Then this paper conducts experiments on the real public dataset of legal judgment prediction. The results show that the proposed method is better than the best model in the legal judgment prediction task. Finally, the development of legal judgment is discussed.

Key words: legalal judgement prediction, multi-layer structure information of law, deep learning

摘要: 近年来,法律领域的智能化引起了学界的广泛关注。选取法律领域中十分重要的法律判决预测任务作为研究重点,法律判决预测包含推荐相关法条、定罪和刑期预测等三个子任务。随着深度学习在各个领域的广泛应用,一些研究者将深度学习方法引入法律判决预测任务并取得了较好的效果。现有基于深度学习的法律判决预测方法通常是通过构建案情描述和法条之间的注意力来提升模型预测能力,或者利用三个法律判决预测子任务间的关系来提升整体的性能。但是这些工作未考虑法律文本中的多层层次化信息,如刑法第三百九十七条包含职务侵占罪和玩忽职守罪,其法条大类是渎职罪,并且每个罪行有不同的刑期。针对该问题,考虑引入法律文本的多层层次化信息用于法律判决预测任务。具体来说,对法律文本的多层结构信息进行预处理,并利用协同注意力机制将法条的多层信息融入到案情描述中,得到每个子任务的融合不同层次的法律信息的案情描述表示,从而提升司法判决预测任务的性能。在真实的法律判决预测任务公开的数据集上进行了实验,结果显示提出的融合法律文本多层结构信息的模型在法律判决预测任务上优于当前最好的模型。对法律智能化的未来和发展进行了展望。

关键词: 法律判决预测, 法律多层结构信息, 深度学习