Beyond Sentiment Polarity: Cognitive Theory-Based Emotion Type Analysis of Social Media Text for Business Applications

Abstract

The emotion types expressed in social media have a strong influence over the perspectives and judgments of readers and their decision making in commercial activities. Developing the capability to automatically and unobtrusively detect and evaluate the expression of emotions in social media is of great interest to consumers, firms, and researchers. In this dissertation, we present a novel framework and methodology for the automated analysis of the emotions expressed in online social media content. Its design is informed by a leading cognitive theory of emotions. Our approach addresses two of the major technical challenges faced by established methods for opinion and emotion analyses. It utilizes semi-supervised learning to bootstrap knowledge on the expression of emotions in a target domain of application, eliminating the requirement for annotated training data; and rule-based classification inspired by the leading cognitive theory of emotions, analysis easily interpreted and relied upon by an end-user. To evaluate our approach to emotion analysis, we conduct a series of empirical experiments on several datasets in emotion type classification and key business applications of emotion analysis predicting several social media measures of interest. The results indicate that our proposed approach accurately identified expressions of emotions in social media and outperformed the established methods in emotion type classification. Emotion information extracted by our proposed approach was more effective in key business applications than sentiment polarity information in predicting social media measures of interest. This study offers novel contributions to the research on emotion analysis and into emotion theory, as well as insights for practitioners on evaluating expressions of emotions in social media and business applications of emotion analysis.