Twitter Likes Prediction Using Content and Link based Features

Abstract

Twitter, a microblogging network, allow its users to post content in real-time according to their interest and share ideas, thoughts and information with each other. Contents can be an image, a movie, a link to a news article or a short message known as “Tweet”. Although Twitter provides a list of most popular topics, called Trending Topics, but users are usually concerned about a small quantity of tweets from their own topic of interest. It is rather challenging to predict which kind of information is expected to attract interest of more users in such a large collection of tweets and can become more popular within short time interval. In this study, we use the “likes” of tweet as a measurement for the popularity among the Twitter users and study the interesting problem of Tweet Likes Count Prediction (TLCP) to explore the characteristics for popularity of tweets for top Trending Topics in the near future. Valuation of possible popularity is of great importance and is quite challenging. For a particular Tweet, we measure the impact of three main attributes (Tweet Content, Number of followers and Geographical Location) for TLCP by using prediction models and evaluate their performance using F-measure. A real world dataset from Twitter was extracted covering tweets from August 4, 2016 till August 21, 2016. Experimental results show that Bayesian Network outperform 70% performance with combined features (Tweet, Followers, Location) on likes as a best predictive model than others on the basis of Accuracy, Precision, Recall and F-measure.