面向裁判文书的行为要素挖掘与分析关键技术研究

针对裁判文书时序性和逻辑性强等特点，面对行为要素的识别困难、时序关系识别缺失、案情结构化语义分析薄弱等问题，本项目引入反绎学习方法解决案情行为要素表达不固定、案情行为要素抽取中训练不足带来的识别困难问题，引入“行为要素链”的概念解决时序关系和语义问题，研究以“行为”为核心的案情行为要素抽取以及语义分析方法。在此基础上，利用“行为要素链”探索案件分析的方法，重点研究法条推荐、类案推送等辅助审判的模型和方法，最后，进行司法辅助审判实践验证。重点解决两个关键技术问题：（1）复杂环境下案件行为要素的识别问题；（2）基于“行为要素链”的案件相似度计算问题。通过本项目的实施，可以实现对裁判文书中的案情语义的深层分析，有效提高司法智能化辅助审判的水平。

In view of the sequential and logical characteristics of adjudicative documents, in the face of the difficulty in identifying behavior elements, lack of time series relation identification and weak semantic analysis of case structure, this project introduces the method of abductive learning to solve the problem of recognition difficulties caused by unfixed expression and the lack of training in the extraction of behavior elements, introduces the concept of "behavioral elements chain" to solve sequential relationship and semantic problems, and studies the extraction of behavioral elements with "behavior" as the core in cases. On this basis, "behavior element chain" is used to explore the calculation method of case analysis, focusing on the model and method of auxiliary trial such as law recommendation, class case push and so on. Finally, the judicial auxiliary trial practice verification is carried out. The project solving two key technical problems: .(1) identification of case behavior elements in complex environments ;(2) case similarity calculation based on the "chain of behavior elements ". Through the implementation of this project, we can realize the deep analysis of the case semantics in the judgment documents, and effectively improve the level of judicial intelligent auxiliary trial.