非线性递推辨识理论在量刑数据分析中的应用  

Applications of nonlinear recursive identification theory in sentencing data analyses

作　　者：王芳[1] 张蓝天 郭雷[2] Fang WANG;Lantian ZHANG;Lei GUO(Data Science Institute,Shandong University,Jinan 250100,China;Academy of Mathematics and Systems Science,Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]山东大学数据科学研究院,济南250100 [2]中国科学院数学与系统科学研究院,北京100190

出　　处：《中国科学：信息科学》2022年第10期1837-1852,共16页Scientia Sinica(Informationis)

基　　金：国家自然科学基金(批准号:12288201)资助项目。

摘　　要：司法大数据已成为法律实证研究和智慧司法工程建设的重要基础,相应地,数据计算结果的可解释性与可靠性等基础性问题愈加重要.为此,我们对非线性递推辨识理论进行了相应创新,并应用于量刑数据分析.具体来讲,依据相关法律建立了更加符合法逻辑的非线性随机量刑模型(S-模型),应用我们提出的非线性递推辨识算法和建立的关于有限数据样本下辨识精度的数学理论,对近20万故意伤害罪判决数据进行了计算分析.研究发现,与传统线性模型及最小二乘算法相比,基于我们的S-模型和非线性递推辨识算法所给出的计算结果,更符合量刑基本原则和具体规则,可以更准确地反映量刑要素的影响及变化,并具有更好的预测能力.Judicial data have become an important base for empirical legal studies and the construction of smart court systems. Thus, the interpretability and reliability of the computation results become increasingly important. Therefore, we established a new theory of nonlinear recursive identification and applied it to sentencing data analyses. Specifically, based on the criminal law and sentencing guidelines, we established a nonlinear stochastic model for sentencing, to which the corresponding nonlinear recursive identification theory is applied with a specified finite length of data, which is nearly 200000 judgment documents of the crime of intentional injury. Compared with the traditional linear regression model and the least squares algorithm, the computation results given by the nonlinear recursive identification theory are more consistent with the basic principles and rules of sentencing, can reflect the effects and changes of sentencing factors more accurately, and exhibit better prediction ability.

关 键 词：非线性模型 递推辨识算法 辨识精度保证 量刑 故意伤害罪 司法判决数据 

分 类 号：D924.1[政治法律—刑法学]