Actionable Pattern Discovery for Tweet Emotions

Abstract

The most popular form of communication over the internet is text. There are wide range of services that allow users to communicate in the natural language using text messages. Twitter is one such popular Micro-blogging platform where users post their thoughts, feeling or opinion on a day-to-day basis. These text messages not only contain information about events, products and others but also the writer’s attitude. This kind of text data is useful to develop systems, which detect user emotions. Emotion detection has wide variety of applications including customer service, public policy making, education, future technology, and psychotherapy. In this work, we use Support Vector Machine classifier model to automatically classify user emotions. We achieve accuracy in the range of 88%. The Emotional information mined from such data is huge and these findings can be more useful if the system is able to provide some actionable recommendations to the user, which help them, achieve their goal and gain benefits. The recommendations or patterns are Actionable if user can perform action using the patterns to their advantage. Action Rules help discover ways to reclassify objects with respect to a specific target, which the user intends to change for their benefits. In this work, we focus on extracting Action Rules with respect to the Emotion class from user tweets. We discover actionable recommendations, which suggests ways to alter the user’s emotion to a better or more positive state.