传统暗黑人格监测方式存在诸多问题，并且针对青少年群体的暗黑人格监测一直没有受到足够的重视。故本研究利用问卷调查法，同时监测多种类型的模态数据，在深圳布吉高级中学对该校高一，高二，高三合计91名学生进行了暗黑人格研究。

本研究利用了问卷调查法，运用了暗黑十二条量表，大五人格问卷，以及道德认同量表；并且监测了多种数据，如被试的血压，面部温度，心率，微表情等；让被试进行自陈式讲述获取其文本数据；再利用道德认同问卷来检测其道德敏感性，最后用金额分配游戏来进行监测被试的与利他倾向。

实验一采取自陈式描述，让被试进行联想式描述，并对被试描述时的词汇进行总结汇总，并及记录其显著的模态信息并建立模型，单一模态的模型准确度达到80%，并利用文本与HRV共同搭建预测系统，准确率达到94%。

实验二让半数高暗黑人格与低暗黑人格被试组成实验组，观看升华感影片，让剩余被试观看无关视频。对比高低暗黑特质被试在观看升华视频时的表现来对暗黑人格尝试自动化监测，并让两组被试观看视频后填写相应问卷以测量其道德敏感性，再采用金钱分配范式，让被试模拟面对四类不同身份但相同贡献合作者进行金钱分配，来监测其利他倾向。验证升华影片能否提高人群的道德情绪并产生利他倾向以及高暗黑特质是否会导致被试在观看升华视频中与普通被试出现显著差异。

结果表明，观看升华视频组相比观看无关视频组分配给他人更多的金额（F（1,81）= 9.839， p = 0.002, Cohen's d = 0.43），并且在道德敏感性得分上得分更高（F（1,81）= 4.499， p = 0.037）。这说明观看升华视频可以显著提升被试的道德情绪与利他倾向。采取交叉验证，验证高低暗黑组在道德情绪提升与利他倾向是否会因为暗黑特质而存在显著差异，结果表明，暗黑特质组与观看升华视频组的交互作用对分配金额没有显著影响（F（1,81）= 0.061， p = 0.806）。且暗黑特质组与观看升华视频组的交互作用对道德敏感性得分也没有显著影响（F（1,81） = 0.279， p = 0.599）。最终总结其并不存在显著的调节或中介效应。同时对被试进行多模态信息的收集与建模，并最终尝试通过提取文本，HRV，表情，面部温度来对被试的暗黑特质来实现自动化监测，单一模态建模准确度最高达到80%，最终利用了面部温度变化，文本，HRV，表情等数据共同搭建机器学习模型，准确率达到95%。

The traditional assessment methods for dark personality have several problems, and darkpersonality monitoring in adolescents has not received enough attention. This study aimed to investigate dark personality using a questionnaire survey and multiple modal data monitoring. It involved 91 students from grades 10 to 12 at Buji Senior High School in Shenzhen. The Dark Triad Scale, Big Five Personality Questionnaire, and Moral Identity Scale were used, and
data such as blood pressure, facial temperature, heart rate, and micro-expressions were monitored. Additionally, the participants provided self-reported text data through free- association descriptions. The Moral Identity Questionnaire was used to measure moral sensitivity, and the money allocation game was used to monitor altruistic tendencies. Experiment 1 utilized free-association descriptions to summarize and gather significant
modal information from the participants' descriptions and built models. The single model accuracy rate was 80%, and a prediction system was established using text and HRV data, with an accuracy rate of 94%. In Experiment 2, high and low dark personality trait participants were divided into experimental groups and watched a transcendent video while the remaining participants watched an unrelated video. The behavior of high and low dark personality trait participants while watching the transcendent video was compared to attempt to automate the monitoring of dark personality traits. Participants then completed a corresponding questionnaire to measure their moral sensitivity and played a money allocation game to measure their altruistic tendencies. The results revealed that the group that watched the transcendent video allocated more money to others compared to the group that watched the unrelated video (F(1,81) =9.839, p = 0.002, Cohen's d = 0.43), and scored higher on moral sensitivity (F(1,81) = 4.499, p = 0.037). This implies that watching a transcendent video can significantly enhance moral emotions and altruistic tendencies. Cross-validation was used to verify whether the high and low dark personality trait groups differed significantly in moral emotions and altruistic tendencies due to their dark personality traits. The results showed no significant moderation or mediation effects between the dark personality trait group and the transcendent video group onthe amount of money allocated (F(1,81) = 0.061, p = 0.806), and the moral sensitivity score (F(1,81) = 0.279, p = 0.599). In conclusion, there were no significant moderation or mediation effects. Multiple modal information was collected and modeled, and an attempt was made to automatically monitor participants' dark personality traits by extracting text, HRV, expressions, facial temperature, etc. The highest accuracy rate of a single modal model reached 80%, and a machine learning model was built using facial temperature changes, text, HRV, expressions, etc., with an accuracy rate of 95%