Abstract: Abstract：With the development of artificial intelligence in deep learning technology in recent years, emotional computing and personality computing technology are maturing day by day and have achieved good performance in many practical application scenarios. Applying artificial intelligence emotional computing technology to the field of crime risk assessment can effectively solve the problem of intra-individual difference prediction factor assessment, which is difficult to be solved by current mainstream risk assessment tools, and the problem of distortion of results caused by social approval of participants being assessed. At present, the mainstream assessment tools are mainly structured clinical assessment tools and statistical actuarial assessment tools. The limitations of structured clinical violence risk assessment tools are mainly reflected in the following aspects: there are explanatory barriers to the inherent causal mechanism between mental disorders and violent crimes, and individuals with mental disorders may not necessarily have high risk of violent crimes; Although the statistical actuarial crime risk assessment tool performs well in prediction validity, it lacks the explanation of causal mechanism between prediction factors and criminal acts, so it is difficult to form a systematic explanation of the causes of criminal acts. A large number of relevant factors are incorporated into the statistical actuarial evaluation tool, which makes the tool more "bloated" and takes a lot of time to complete an evaluation. Most statistical actuarial evaluation tools obtain data through self-evaluation and self-presentation, which are easily affected by social approval effect and distort the data. The mainstream crime risk assessment tools are mainly applied to the field of crime risk assessment in prisons and re-crime risk assessment after leaving prison. Due to the long time-consuming operation of static assessment tools, the unwillingness of criminals in prisons to cooperate with the assessment work, and the serious deviation of assessment results from methods such as social approval effect, the static assessment tools are confronted with realistic difficulties such as assessment distortion, time-consuming and labor-consuming, and difficulty in realizing dynamic real-time crime risk assessment in prisons. In order to solve the above practical difficulties, a dynamic crime risk assessment tool based on emotional psychology and supported by artificial intelligence emotional computing technology has been developed, which focuses on solving the problems of violence risk and escape risk assessment that are the focus of crime risk assessment in prisons. The crime risk dynamic assessment tool is based on emotional psychology and supported by artificial intelligence emotional computing technology. It objectively observes and calculates the emotional time series state of the assessment object, and automatically obtains the emotional type and emotional fluctuation degree of the assessment object at a certain point in time. The dynamic criminal risk assessment tool based on emotional computing technology can solve the problem of risk factor assessment and prediction of intra-individual differences, which is difficult to solve with previous tools. At the same time, because emotion computing technology mainly uses cameras to collect individual image data, it does not need appraisers to answer questions, which can effectively avoid the appearance of social approval effect. The future development of the crime risk dynamic assessment tool is mainly to realize the echo with the static assessment tool to realize the measurement technology.

Key words: Key words：crime risk assessment, Emotional computing, Dynamic evaluation

摘要： 摘 要 随着近年来人工智能深度学习技术的发展，情感计算与人格计算技术日渐成熟，在许多实际应用场景中取得了良好的表现，当前人工智能情感计算技术应用于犯罪风险评估领域，能够有效解决目前主流的风险评估工具难以解决的个体内差异性的预测因子评估问题以及被测评参与者因社会赞许性而导致结果失真的问题。本文在详细阐述目前主流评估工具的局限性基础上，详细阐述了以情感计算技术为支撑的动态风险评估工具的设计思路、目前已有的技术方案以及设计细节的理论依据，在此基础上最后提出以人工智能技术为支撑的新型评估工具的未来发展方向。

关键词: 关键词 犯罪风险评估, 情感计算, 动态评估