Automatic Classification of Social Media Messaging Using Multi-Dimensional Sentiment Analysis and Crowdsourcing

Abstract

In order to study the use of Twitter, it would be useful to be able to easily classify large volumes of messaging being used by Twitter users besides just in the traditional dimensions of positive and negative sentiment. For instance, in the space of social media marketing, it would be useful to be able to automatically identify social media messages into the classic framework of informative, persuasive and transformative advertising. In this paper we propose a general method for using crowdsourced training data labels for supervised machine learning, in order to automatically classify social media messaging with multidimensional sentiment analysis. We proceed with our methodology to create classifiers in the context of a firm’s messaging, but we hypothesize that its applications are more general than this context. Our results show that the proposed method works for the relatively objective attribute of informativeness, but is not as useful for more subjective attributes such as persuasiveness and transformativeness. We finish by discussing recommendations for future work in this area.

Keywords: Social Media, Machine Learning, Sentiment Analysis, Informativity, Transformativity, Persuasiveness, Advertising

JEL Classification: M37, C80