A two-phase approach using LDA for effective domain-specific tweets conveying sentiments

Abstract

Twitter is a free social networking platform where people can post and interact with short messages known as "Tweets". The freedom of being able to reach out to the world in a fraction of seconds has made Twitter an effective medium for the general public to express their opinion on a global scale. Since Tweets have the potential to make a global impact, companies too have started using the service to reach out to their customers. Moreover, in spite of this service being immensely effective, it is found challenging by many users to express their views through a Tweet due to the restriction imposed of minimum 280 characters. The proposed work is aimed at helping people compose better quality Tweets belonging to a specific domain in the restricted character limit. The system is designed to mine important features/topics about a domain using Latent Dirichlet Allocation (LDA) algorithm and to compute the polarity of the sentiment words associated with them with respect to the domain using a two-phase approach on an Amazon review corpus. The discovered topics/features and sentiments are recommended as suggestions to Twitter users while composing new Tweets. The paper describes and presents initial results of the system on cell phones and related accessories domain.