Combining Textual Cues with Social Clues: Utilizing Social Features to Improve Sentiment Analysis in Social Media

Traditional sentiment analysis methods do not perform well when applied to social media data. In this study, we propose an approach to improve sentiment analysis performance in the context of social media. Our approach utilizes three types of additional information that can be collected from social media platforms-personal preference, friend influence, and herding effect-to enrich the input features of a supervised sentiment classification model. We implement the approach on data sets collected from Twitter across two industries (airlines and wireless service providers) and present the performance improvement attained by combining social features with pure text-based features. To further investigate the operational implications of this improvement, we develop a stylized service recovery model for customer relationship management in social media. Our work has implications for automating social media monitoring and, more broadly, for improving customer relationship management in organizations.