联合多源分析的罪名预测研究

Research on Charge Prediction Based on Multi-Source Joint Analysis

随着人工智能技术逐步在司法领域落地与应用,法律人工智能的研究成果极大地提升了司法从业人员的工作效率。罪名预测作为法律人工智能的重要核心应用之一,旨在根据案件描述预测犯罪主体触犯的刑法罪名。针对目前罪名预测研究仅依赖于单一的裁判文书数据源,但裁判文书对案件细节的阐述不够全面的问题,构建了一个结合裁判文书和庭审文书的多源联合分析数据集,将概述内容的裁判文书与囊括细节的庭审文书相结合进行罪名预测。在构建的多源联合数据集上进行了大量实验及分析,实验结果验证了裁判文书和庭审文书在信息上的互补性,为罪名预测任务提供了新的思考角度。

With the gradual landing and application of artificial intelligence technology in the judicial field,the research results of legal artificial intelligence have greatly improved the work efficiency of judicial practitioners.As one of the important core applications of legal artificial intelligence,charge prediction is aimed at predicting criminal charges committed by criminal subjects based on case descriptions.Aiming at the problem that the current charge prediction research only relies on a single data source of the judgment document,but the judgment document does not fully explain the details of the case,this paper constructs a multi-source joint analysis dataset that combines the judgment document and the court trial document.The research combines the judgment document summarizing the content with the court trial document covering the details to predict the crime.This paper conducts a lot of experiments and analyses on the constructed multisource joint dataset.The experimental results verify the complementarity of information between the judgment document and the court trial document,and provide a new thinking angle for the charge prediction task.