在线主动自杀预防的群体识别及其特征的研究
文献类型：学位论文
| 作者 | 黄彦 |
| 答辩日期 | 2021-01 |
| 文献子类 | 硕士 |
| 授予单位 | 中国科学院心理研究所 |
| 授予地点 | 中国科学院心理研究所 |
| 其他责任者 | 朱廷劭 |
| 关键词 | 自杀 群体 预防 识别 特征 |
| 学位名称 | 理学硕士（同等学力硕士） |
| 学位专业 | 健康心理学 |
| 其他题名 | Research on Group Identification and Characteristics of Online Proactive Suicide Prevention |
| 中文摘要 | From the perspective of today's mobile social trends, more and more people choose to share life on the network platform and express their feelings. Social media is not only a tool for modern people to express themselves, communicate and participate in group activities, but also a window to reflect their psychological state, and an important channel for implementing online suicide prevention. Sina Weibo is the most representative social media in China. The user's posting data on Sina Weibo contains language features related to suicidal behavior. Extracting the language features presented by these user groups when using Weibo can establish a suicide risk prediction model, automatically identify users with suicidal thoughts, and further carry out online intervention for these potential suicide groups.Based on the existing research results and Sina Weibo's online active suicide prevention process, this topic proposed an optimized idea for group identification of suicidal ideation, and conducted a study on the establishment and performance of suicide recognition models. The results show that: The suicidal idea recognition model based on multi-feature weighting method (MFWF method), its precision, recall, F-measure and accuracy reached 0.89, 0.88, 0.88, 0.89, which are higher than data-driven method (p<0.01). It is an improvement to the data-driven method of single feature extraction, and its recognition effect is significantly better than the existing data-driven methods.In the study of group intervention of suicidal ideation, word frequency analysis was performed on the content of the user's replies. The results show that 10 words including "thank you", "doctor", "emotion" and "depression" are high-frequency words, and the high frequency use of mental disease-related nouns indicates that users who participate in online suicide interventions appear significantly pay attention to depression and related clinical features, or they were plagued by persistent depression and pessimism, which leads to specific questions, anxieties and discussions. Thus, the recognition model could detect the potential high-risk suicide users with depression and depressive symptoms.In this paper, the content of each user's reply message is recorded according to certain classification standards. After the classification of 3626 suicidal users, 3284 people can be identified. In general, 90.6% of the users' replies confirmed that they had different degrees of suicidal ideation when they published information, which shows that the accuracy of the recognition model is good. At the same time, this study also builds a classification basis for further understanding of different users in the suicidal ideation group.In the intervention research of high self-exposure (HSD) users, statistics show that psychological factors, social factors and intimacy factors are the three main suicide ideation sources, specifically to the segmentation factors, which are mental illness, social pressure and parent-child problems. These three factors appear in a dual way, which should be paid close attention in suicide intervention. In the statistics of mental diseases for HSD users, the number of users with depression and schizophrenia accounts for 65.9% of the total group, and the number of users with depression was the largest, with a case ratio of 84.5%. This further illustrates that most users with high self-exposure are indeed at high risk of suicide, and they have obvious group characteristics in the source of suicide ideation and the distribution of mental diseases.The results of this study not only provide us a better method for screening the groups with suicidal ideation, but also facilitate targeted support with the help of in-depth understanding about group characteristics, and help to establish a more systematic and reasonable online work flow of proactive suicide prevention from identification to intervention. |
| 英文摘要 | 从当今的移动社交趋向来看，越来越多的人选择在网络平台上分享生活，表达自己的感受。社交媒体不仅是现代人自我表达、交流、参与群体活动的工具，也是反映他们心理状态的窗口，是实施在线自杀预防的一个重要渠道。新浪微博是我国较具代表性的社交媒体。在新浪微博上用户的发布数据，包含了与自杀行为相关的语言特征。提取这些用户群体在使用微博时所呈现的语言特征，可以建立自杀风险预测模型，且能自动化识别出有自杀意念的平台用户，进一步对潜在的自杀意念群体开展在线干预的工作。本课题在已有的研究成果基础上，基于新浪微博的在线主动自杀预防流程，提出对自杀意念群体识别的优化思路，并就此进行自杀识别模型的建立及性能研究，结果显示：本研究提出的自杀意念识别方法——多特征融合加权法（MFWF法），其精确率、召回率、F值和准确率分别达到0.89、0.88、0.88、0.89，均高于数据驱动方法（p<0.01），是对单特征提取数据驱动方法的改进，其识别效果明显优于现有的数据驱动方法。在自杀意念群体干预的研究中，对用户的回复内容进行词频分析。结果显示，“谢谢”、“医生”、“情绪”“抑郁症”、“抑郁”等10个词语为高频用词，精神疾病相关名词的高使用频率表明了参与在线自杀干预的用户，明确出现了对抑郁症及相关临床特征的关注，或因情绪持续低沉、悲观而困扰，因此产生了特定的疑问、焦虑和讨论。由此可见，识别模型较好地检出了患有抑郁症及有抑郁症状的潜在高危自杀用户。本课题将每个产生回复用户的私信内容，按照一定的分类和相应标准形成标注记录，经过对3626名自杀意念用户进行分类，统计得出3284人为可确认对象。总体而言，90.6%的用户回复内容确认了其在发布信息时确有不同程度的自杀意念，可见识别模型的准确度较好，同时该研究也为进一步了解自杀意念群体中的不同用户构建了分类基础。在高自我暴露（HSD）用户的干预研究中，统计显示，心理因素、社会因素和亲密关系因素是用户最主要的三大自杀意念来源，具体到细分因素，是精神疾病、社会压力和亲子问题，三者存在两两共同出现的现象，应该在自杀干预工作中予以重点关注。在针对HSD用户的精神疾病统计中，患有抑郁症与精神分裂症患者的用户数占总人数的65.9%，患有抑郁症的人数最多，个案比达到84.5%。这进一步说明了，高自我暴露的用户大部分确为高危自杀者，在自杀意念来源与精神疾病的分布上，他们表现出了较明显的群体特征。本课题的研究结果，不但为我们筛查自杀意念群体提供更好的方法，而且对于群体特征的深入了解，也利于向他们提供更有针对性的干预支持，总体从识别到干预建立起更系统、更合理的在线主动自杀预防的工作流程。 |
| 语种 | 中文 |
| 源URL | [http://ir.psych.ac.cn/handle/311026/41681]  |
| 专题 | 心理研究所_社会与工程心理学研究室 |
| 推荐引用方式 GB/T 7714 | 黄彦. 在线主动自杀预防的群体识别及其特征的研究[D]. 中国科学院心理研究所. 中国科学院心理研究所. 2021. |
入库方式： OAI收割
来源：心理研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。