Mutual Interest-Based Twitter Followee Recommendation Using Latent Dirichlet Allocation Topic Modelling

Twitter is a social media micro-blogging platform which is most actively and widely used worldwide. Tweets give almost real-time updates on anything and everything that concerns a Twitter user. It may be product or service reviews, events, political opinions, weather updates, breaking news, etc. Twitter has a mechanism to subscribe to the tweet updates of selected users by ‘following’ them. This leads to the idea of using a recommendation engine that will suggest suitable ‘followee’ (whom to follow) that matches user preferences and choices most. Existing recommendation algorithms are mainly based on phone or email contacts and location. Some recent works analyzed tweets to suggest the persons to ‘follow’. The Latent Dirichlet Allocation (LDA) probabilistic topic model is popularly used to cluster the tweet texts. In this work, we enhance this concept further by considering the recentness of peoples’ interest in the recommendation process. A framework for the Twitter followee recommendation system is proposed based on the temporal interest of the user using LDA topic modeling. Tweets of existing users are clustered with LDA and cosine similarity is used to derive the recommended user list. The proposed system is implemented and executed with various input parameters on tweets collected through Twitter stream API and the results were validated manually to evaluate the performance of the system.