A Data-driven Approach to Dynamically Learn Focused Lexicons for Recognizing Emotions in Social Network Streams

Opinion Mining aims at identifying and classifying subjective information in a collection of documents. A variety of approach exists in literature, ranging from Supervised Learning to Unsupervised Learning. Currently, one of the biggest opinion resource of opinionated texts existing on the Web is represented by Social Networks. Networks are not only a vast collection of documents but they also represent a dynamic evolving resource as the users keep posting their own opinions. We based our work relying on this idea of dynamicity, building an evolving model that updates itself in real time as users submit their posts. This is done through a set of supervised techniques based on a Lexi- con of emotionally-tagged terms (i.e. anger, disgust, fear, joy, sadness and surprise) that expands accordingly to user's dynamic content.