Abstract
Live-stream suicide has become an emerging public health problem in many countries. Regular users are often the first to witness and respond to such suicides, emphasizing their impact on the success of crisis intervention. In order to reduce the likelihood of suicide deaths, this paper aims to use psycholinguistic analysis methods to facilitate automatic detection of negative expressions in responses to live-stream suicides on social media. In this paper, a total of 7212 comments posted on suicide-related messages were collected and analyzed. First, a content analysis was performed to investigate the nature of each comment (negative or not). Second, the simplified Chinese version of the LIWC software was used to extract 75 psycholinguistic features from each comment. Third, based on 19 selected key features, four classification models were established to differentiate between comments with and without negative expressions. Results showed that 19.55% of 7212 comments were recognized as "making negative responses". Among the four classification models, the highest values of Precision, Recall, F-Measure, and Screening Efficacy reached 69.8%, 85.9%, 72.9%, and 47.1%, respectively. This paper confirms the need for campaigns to reduce negative responses to live-stream suicides and support the use of psycholinguistic analysis methods to improve suicide prevention efforts.