Mining Emotions of the Public from Social Media for Enhancing Corporate Credit Rating

The proliferation of online social media has been changing the ways how individuals interact with corporations. Previous studies have examined how to extract investors' sentiments captured on social media to enhance stock prediction. However, little work was done to leverage public's emotions captured on social media to predict corporate credit risks. Our research fills the current research gap by developing a new computational method to extract public's emotions embedded in social postings to supplement common financial indicators (e.g., return-on-assets) for predicting corporate credit ratings. Grounded in Plutchik's Wheel of Emotions, the proposed computational framework can automatically extract the distribution of eight basic emotions from textual postings on online social media. In particular, one main contribution of our work is the development of the new emotion latent dirichlet allocation (ELDA) model for textual emotion analysis. In addition, we develop an ensemble learning model with random forest (RF) as the basis classifier to improve the performance of corporate credit rating. Based on the real-world data crawled from Twitter, our experimental results confirm that the proposed ELDA model can effectively and efficiently extract public's emotions from social postings to enhance the prediction of corporate credit ratings. To our best knowledge, this is the first successful research of developing a new computational model of extracting public's emotions from social postings to enhance corporate credit risk prediction.