Affect Proxies and Ontological Change: A Finance Case Study

Traditional sentiment analysis has been focusing on inference of the sentiment polarity using sentiment-bearing words. In this paper, we propose a new way of studying sentiment and capturing ontological changes in a domain specific context in the perspective of computational linguistics using affect proxies. We used Nexis service to create a domain specific corpus focusing on banking sectors. We then created an affect dictionary from three kinds of lexica: sentiment lexica as in the General Inquirer dictionary; news flow represented by domain entities such as financial regulators and banks; and what we call contested term lexica, which consists of terms whose semantic implication is inconsistent over time. Univariate and multivariate analysis techniques such as factor analysis are used to explore the relationships and underlying patterns among the three types of lexica. Analysis results suggest that citations of regulatory entities show strong correlation with negative sentiments in the banking context. Also, a factor analysis was conducted, which reveals several groups of variables in which the contested terms correlate with positive and negative sentiments.