Social media text mining for detecting behavioral and psychological conditions in children

点击翻译按钮获取中文摘要

The widespread usage of social media by children has recently raised problems related to cyberbullying andother threats to their well-being. VISR.CO has an app that alerts parents to issues their kids are facing online.They analyze the social interactions kids are experiencing on social networks, and search for over 25 alertcategories, from bullying to mental health. Highly accurately detecting these issues that parents care about is adifficult Machine Learning (ML) and Natural Language Processing (NLP) challenge, requiring the latestadvancements in computer science. Additionally, when detecting mental health issues, such as anxiety,depression and self-harm, a clinical expertise is required to understand the underlying conditions and causes.They are currently able to deliver very rudimentary mental health alerts but require both NLP expertise andclinical insight.As of February 12, 2016, VISR.CO has been actively monitoring 9,985 children on behalf of 8,645 parents.VISR.CO has uncovered 850 confirmed cases of bullying (24% success rate), 452 confirmed mental healthconcerns (21% success rate), and 1,018 confirmed substance use concerns (16% success rate). We propose toresearch ways to increase our success rates by using more advanced data-mining techniques. Another challengethat they need to address is to find better ways to deal with the social media texts (such as frequentmisspellings, short-lived idiomatic expressions, and abbreviations) and to account for the dynamic nature oflinguistic expression across social media platforms.If these challenges are addressed, the product will have a competitive advantage by staying up to date,delivering useful alerts to parents, and intelligently framing the unique issues found the online activity of everychild.