逮捕审查判断中犯罪嫌疑人社会危险性的量化评估

通过对2020年以来13万余份起诉书和不起诉书的机器学习,研究发现:我国司法人员在判断羁押必要性时,会评估犯罪嫌疑人的社会危险性,但相应评估侧重于罪行危险性因素,却忽略了人身危险性因素;引入量化评估方法,主要有利于改善仅依靠基本案情信息难以准确判断是否需要羁押的“复杂”案件的羁押必要性判断,量化方法有可能显著降低此类案件的羁押率。当前在判断羁押必要性时,办案人员主要依靠罪行危险性因素的主观综合判断,规范改革路径着力于规则细化和要件重构,但这无力化解社会危险性判断信息不足的问题。引入大数据建模方法,有助于系统统合零散的社会危险性判断信息。此外,可通过问卷、量表、数字化设备等方式扩充模型的训练数据,通过深化对社会危险性发生机制的理解,区分案由和社会危险性类型,构建多个子模型,以进一步提升量化工具的准确性。

Through machine learning on more than 130,000 legal documents since 2020,this study reveals that when making detention decisions,Chinese judicial personnels assess the social dangerousness of criminal suspects,but such assessments focus on elements of“criminal dangerousness”while overlooking those of“personal dangerousness”.The introduction of the quantitative assessment method is mainly beneficial for the handling of“complex”cases in which it is difficult to accurately determine whether to detain the suspect relying solely on basic case informations.It is argued that the quantitative method has the potential to significantly reduce the detention rate in such cases.In the current practice of detention decisions,case handlers mainly rely on subjective assessment based on the elements of“criminal dangerousness”.Traditional reform path focuses on rule refinement and element reconstruction,but falls short of addressing the insufficient assessment of social dangerousness.Incorporating big data modeling methods can help to systematically integrate scattered detention decision information.In addition,we can also expand the training data of the model by using methods such as questionnaires,scales,and digital devices,and further improve the accuracy of quantitative tools by deepening the understanding of the mechanisms underlying social dangerousness and distinguishing between causes and types of danger to construct multiple sub-models.