Substance use and sentiment and topical tendencies: a study using social media conversations of youth experiencing homelessness
本研究分析无家可归青少年在Facebook上的对话，发现情感倾向和话题（如金钱）与物质使用（如大麻）相关，展示了利用社交媒体数据补充传统调查的可行性。
Purpose This study investigates associations between Facebook (FB) conversations and self-reports of substance use among youth experiencing homelessness (YEH). YEH engage in high rates of substance use and are often difficult to reach, for both research and interventions. Social media sites provide rich digital trace data for observing the social context of YEH's health behaviors. The authors aim to investigate the feasibility of using these big data and text mining techniques as a supplement to self-report surveys in detecting and understanding YEH attitudes and engagement in substance use. Design/methodology/approach Participants took a self-report survey in addition to providing consent for researchers to download their Facebook feed data retrospectively. The authors collected survey responses from 92 participants and retrieved 33,204 textual Facebook conversations. The authors performed text mining analysis and statistical analysis including ANOVA and logistic regression to examine the relationship between YEH's Facebook conversations and their substance use. Findings Facebook posts of YEH have a moderately positive sentiment. YEH substance users and non-users differed in their Facebook posts regarding: (1) overall sentiment and (2) topics discussed. Logistic regressions show that more positive sentiment in a respondent's FB conversation suggests a lower likelihood of marijuana usage. On the other hand, discussing money-related topics in the conversation increases YEH's likelihood of marijuana use. Originality/value Digital trace data on social media sites represent a vast source of ecological data. This study demonstrates the feasibility of using such data from a hard-to-reach population to gain unique insights into YEH's health behaviors. The authors provide a text-mining-based toolkit for analyzing social media data for interpretation by experts from a variety of domains.