Mining emotion-aware sequential rules at user-level from micro-blogs

Social Media have enabled users to keep inter-personal relationships, but also to voice personal sensations, emotions and feelings. The recent literature reports on the potential of technologies based on emotion detection and analysis. However, the understanding of user generated emotional content is a challenging task because it requires the extraction of textual units of interest and the search for potential knowledge nuggets, such as those on the correlation between emotions conveyed over time. In this paper, we study this array of problems through the discovery of structured information on the emotions, which is more difficult than the mere recognition of individual mentions. We propose a framework to discover forms of implication between emotions through high-utility sequential rules. Apart from being emotion-aware and time-aware, these rules have the ability to handle numeric information concerning the quantities of expressed emotions, contrary to the classical association rules designed only for binary data. The application on micro-blogs concerning politics shows the viability of the framework to real-world scenarios and its potential to capture user-level emotional behaviours.