Negative Purchase Intent Identification in Twitter

Abstract

Social network users often express their discontent with a product or a service from a company on social media. Such a reaction is more pronounced in the aftermath of a corporate scandal such as a corruption scandal or food poisoning in a chain restaurant. In our work, we focus on identifying negative purchase intent in a tweet, i.e. the intent of a user of not purchasing any product or consuming any service from a company. We develop a binary classifier for such a task, which consists of a generalization of logistic regression leveraging the locality of purchase intent in posts from Twitter. We conduct an extensive experimental evaluation against state-of-the-art approaches on a large collection of tweets, showing the effectiveness of our approach in terms of F1 score. We also provide some preliminary results on which kinds of corporate scandals might affect the purchase intent of customers the most.