北京大学学报（自然科学版） ›› 2026, Vol. 62 ›› Issue (2): 275-285.DOI: 10.13209/j.0479-8023.2025.098

上一篇    下一篇

基于多源知识聚合的量刑规范性预测方法

景海峰1,2, 王东升3,4,†   

  1. 1. 北京大学软件与微电子学院, 北京 100871 2. 中国石油大学(华东)青岛软件学院计算机科学与技术学院, 青岛 266580 3. 中国政法大学法治信息管理学院, 北京 102249 4. 中国政法大学证据科学教育部重点实验室, 北京 100088
  • 收稿日期:2025-03-04 修回日期:2025-08-05 出版日期:2026-03-20 发布日期:2026-03-20
  • 基金资助:
    中国政法大学科研创新项目(24KYGH013)资助 

A Sentencing Normativity Prediction Method Based on Multi-Source Knowledge Aggregation

JING Haifeng1,2, WANG Dongsheng3,4,†   

  1. 1. School of Software & Microelectronics, Peking University, Beijing 100871 2. College of Computer Science and Technology, China University of Petroleum (East China) Qingdao Institute of Software, Qingdao 266580 3. School of Information Management for Law, China University of Political Science and Law, Beijing 102249 4. Key Laboratory of Evidence Law and Forensic Science, Ministry of Education, China University of Political Science and Law, Beijing 100088
  • Received:2025-03-04 Revised:2025-08-05 Online:2026-03-20 Published:2026-03-20

PDF

摘要：

为辅助审查案件量刑是否规范, 判断量刑属于畸轻、合理或畸重, 构建两个量刑规范性预测数据集, 同时提出一种基于多源知识聚合的量刑规范性预测模型。该模型利用大语言模型, 将刑事案件表示为知识图谱, 通过知识图谱生成案件的类案。为聚合刑法和类案知识, 在模型中设计多源注意力模块, 将该模块生成的多源特征与案件表示一起用于量刑规范性预测。量刑规范性预测对比实验结果表明, 所提模型的F1值高于其他对比模型。消融实验和案例分析表明, 刑法和类案对量刑规范性预测有重要辅助作用。

关键词: 量刑规范性, 多源知识聚合, 类案检索, 注意力机制

Abstract:

To assist in reviewing the standardization of sentencing in cases and determining whether a sentence is disproportionately light, reasonable, or disproportionately heavy, two datasets for predicting sentencing standardization are constructed, and a multi-source knowledge aggregation-based model for sentencing standardization prediction is proposed. The model employs a large language model to represent criminal cases as knowledge graphs, from which similar cases are generated. To integrate knowledge from criminal law and similar cases, a multi-source attention module is designed in the model. The multi-source features generated by this module, together with case representations, are used for sentencing standardization prediction. Comparative experimental results on sentencing standardization prediction show that the F1-score of the proposed model is higher than that of other comparative models. Ablation experiments and case analyses demonstrate that criminal law and similar cases play an important auxiliary role in sentencing standardization prediction. 

Key words: sentencing normativity, multi-source knowledge aggregation, similar case retrieval, attention mechanism

版权所有 © 《北京大学学报（自然科学版）》编辑部
技术支持：北京玛格泰克科技发展有限公司