Fast learning for sentiment analysis on bullying

ABSTRACTBullying is a serious national health issue among adolescents. Social media offers a new opportunity to study bullying in both physical and cyber worlds. Sentiment analysis has the potential to identify victims who pose high risk to themselves or others, and to enhance the scientific understanding of bullying overall. We identify seven emotions common in bullying. While some of the emotions are well-studied before, others are non-standard in the sentiment analysis literature. We propose a fast training procedure to recognize these emotions without explicitly producing a conventional labeled training dataset. We apply our procedure to social media posts on bullying and discuss our findings.