@Article{csse.2019.34.065, AUTHOR = {Yuanyuan Xue, Qi Li, TongWu, LingFeng, Liang Zhao, FengYu}, TITLE = {Incorporating Stress Status in Suicide Detection through Microblog}, JOURNAL = {Computer Systems Science and Engineering}, VOLUME = {34}, YEAR = {2019}, NUMBER = {2}, PAGES = {65--78}, URL = {http://www.techscience.com/csse/v34n2/40028}, ISSN = {}, ABSTRACT = {Suicide has been a perplexing social problem around the world for a long time. Timely sensing hidden suicide risk and offering effective intervention are highly desirable and valuable for individuals and their families. Psychological studies prove that stress status, suicide-related expressions, and social engagement are reliable predictors of suicide risk. However, existing clinical diagnosis can only provide effective treatments to a restricted number of people because of its limited capacity. With the popular usage of social media like microblogs, a new channel to touch the inner world of many potential suicides arises. In this paper, we explore to automatically detect individual’s suicide risk via a microblog platform. Referring to psychology theories, we take one’s stress, self-concerns, suicide-related expressions, last words, social interaction, and emotional traits throughout the posting period on microblogs into account, and construct a 6-dimensional microblog feature space. We examine the differences of these features between the suicide group and the non-suicide group, through a set of real on line blogs posted by those who committed suicide and those who have no suicide intention. The observations reveal the same tendency as psychological theories suggested. To seek the causal relationship between these features and suicide risk, we describe a fuzzy cognitive map (FCM) classification model for suicide risk detection. We test the performance of the FCM classification model on a set of suicide and non-suicide users’ real blogs from the Sina Weibo. The results show that the proposed model is effective and efficient on detecting users’ suicide risk through Micro-blog, and yields better performance than other machine learning algorithms on small data set, with precision, recall and F1-measure increased by 9.7%, 15.8% and 13% respectively over second algorithm. The results also reveal stress feature vector plays more important role than other feature vectors and can effectively improve the performance of suicide risk detection.}, DOI = {10.32604/csse.2019.34.065} }