Capturing signals of enthusiasm and support towards social issues from twitter

Social media enables organizations to learn what users say about their products online, and to engage with their potential audiences. Social media has also been allowing individual users and the public to signal their enthusiasm, support, or lack thereof for a broad range of topics. In this paper, we analyze the robustness of a prior framework for tagging tweets across the dimensions of enthusiasm (labels: enthusiastic, passive) and support (labels: supportive, non-supportive). We investigate the quality of annotations in a collection of tweets about three topics, namely, cyberbullying, LGBT rights, and Chronic Traumatic Encephalopathy (CTE) in the National Football League. We train models that achieve >70% and 80% F1 score for classifying tweets for enthusiasm and support, respectively. We assess how text-based signals of enthusiasm and support vary depending on the different annotators. Finally, we propose and demonstrate a network analysis-based approach for combining the annotated tweets with account and hashtag mention networks. This step helps to identify top accounts and hashtags related to the considered categories (enthusiasm and support). Our work offers an alternative or supplemental classification schema and prediction model to standard sentiment analysis and stance detection.