TWEET RECOMMENDER MODEL USING ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM

Abstract

Twitter is a ubiquitous, socially engaging and rapid communication medium. To filter the relevant information (news/hashtags/links/follow/tweet) for better user experience recommender systems have been extensively used on Twitter. Uncertainty in user preference, fuzziness in the rating process and the imprecision associated with the voluminous and varied twitter data are some of the difficulties associated which impede enhanced recommendations. This research put forwards an adaptive neuro-fuzzy inference system based tweet recommender model to handle the uncertainty, impreciseness and vagueness in item features and user‟s behaviour. The proposed hybrid (content-based and collaborative filtering based) model learns the interests of users (source tweet user and target tweet user) to categorize tweets. The users are characterized as source user (the user who posted the original tweet) and target user (to whom the tweet is to be recommended). The interests of the source and target user are extracted and the correlation between user interests is established which along with the category of the target tweet are then used to build the neuro-fuzzy model. The results show that the proposed model predicts the recommendation score correctly most of the time with the root mean square error of 0.93. -parameters