Detecting bullying on Twitter using emotion lexicons
Patch, Jerrad Arthur
MetadataShow full item record
Bullying is summarized as unwanted acts of aggression that are likely to be repeated and is difficult to detect through traditional means. This work explored bullying by graphing emotions in Twitter relationships. We performed sentiment analysis by using dictionaries to assign emotions to tweets. We graphed the result as the emotions that were sent from one user to another. We repeated the process over time to extract the user’s emotional relationships. We then had evaluators classify relationships with multiple conversations and tweets for bullying. We then performed a comparison between classifiers using the difference between emotions commonly and uncommonly found in bullying, the text based training set, and the emotion training set (derived from the classified relationships). From this comparison, we found that using the emotional vectors did not improve the accuracy of the classification (78% versus 75%).