基于信息融合的自杀高危人群早期识别关键技术研究与应用

Early detection of at-risk population for suicide is extremely critical for suicide prevention. This study aims to solve the issues of how to promptly, comprehensively and presisely detect and assess the suicide risk, which is the methodological challenge confronted in suicide risk detection. Firsly,text mining techniques and related mathematical algorithms will be used to analyze the last words of suicide victims, blog of community individuals, suicide attempters and suicide completers, to establish database of text indicating suicide risk and related text mining technique. Secondly, survey by questionnaires and semi-structured interview will be employed for data collection for the targeted subjects from several informants and targets themselves, and then Bayesian-networks and other theories will be applied to establish multi-source information fusion techniques for solving the data analysis difficulties during for proxy data for suicide victims from their informants and muliti-source information combning proxy data and self-reported information. Finally, with the techniques established above,integrated suicide risk assessment approaching was established, using multi-source data from people with or without suicidal intention, suicide attempters and suicide completers.Through this study, the authors expect to find ways for fast detection of suicide risk by text, comprehensive and reasonable statistical methods for inconsistency of multi-source information, and intergrated risk assessment during the co-exsistance of risk and protective factors, and lay solid foundations for early detection and accurate assessment of suicide risk for at-risk population.

自杀高危人群的早期识别，对预防自杀意义重大。针对自杀风险识别所面临的"如何快速、全面、准确地识别和评估自杀风险"等方法学难题：本研究首先运用文本挖掘技术及相关算法，利用自杀死亡者遗言、一般人群和自杀者的博客等文字信息，建立自杀风险文本特征库及识别技术；然后，采用问卷调查和半结构式访谈，并通过多位信息人和目标人本人，收集研究对象信息，并运用贝叶斯网络等工具，建立多源信息融合技术，用于解决"多位信息人提供的死者信息"和"兼有代理信息与自我报告信息的多源信息"等的数据分析方法难题；最后，将上述文本识别和信息融合技术，运用于"无自杀意念者、有自杀意念者、自杀未遂者和自杀死亡者"的多源信息收集和分析，建立自杀风险综合评估技术。通过本研究，有望解决"自杀风险文字快速识别"、"多源分歧信息的科学全面分析"、"危险和保护因素共存时的风险综合评估"，为自杀高危人群的早期识别和准确评估提供关键技术和适宜技术。

自杀，是我国目前面临的重要公共卫生问题和社会问题；自杀高危人群的早期识别，对预防自杀意义重大，而目前自杀风险识别面临着“如何快速、全面、准确地识别和评估自杀风险”这一难题。.本研究首先通过网络收集了60位自杀者遗书和1200位一般人群网络日志，进行了文字内容和特征的比较；并结合文献综述，还对25位专家进行了咨询，最终建立了包括226个基础特征词和312个扩展特征词的自杀风险文本特征库。结果发现：以自杀风险文本特征基础词为依据的自杀风险评估技术最佳，其风险评估阈值为27.5,灵敏度为56.7%，特异度为89.2%。然后，募集了416位目标人并进行调查，每位目标人访谈2个信息人，收集自杀相关危险因素信息，进行多源信息融合技术探索和比较。结果发现：IF4、IF5、1st信息融合技术效果相对较佳。最后，研究和比较自杀死亡、自杀未遂和自杀意念的各种危险因素，重点参考自杀行为（包括自杀死亡和自杀未遂）的OR值，结合自杀意念研究结果和文献综述，制定了自杀风险评估技术框架和参考值，该评估技术共纳入10类风险评估因素，包括个人史、家族史、情绪和幻觉、近期负性生活事件、孤独和社会支持，等。.通过本研究，解决了“自杀风险文字快速识别”、“多源分歧信息的科学全面分析”、“危险和保护因素共存时的风险综合评估”，为自杀高危人群的早期识别和准确评估提供关键技术和适宜技术。