Predicting the Best Tweet Using Machine Learning

Abstract

Twitter is an online social networking service that enables users to post and interact with messages known as tweets which are limited to 140 characters. Most of the communication here is done through following or retweeting. Retweet, is a main way to spread information in twitter. It is changing the geography of communication. Recently research focuses on analyzing the factors of retweet behavior. It's data feed includes all kinds of meta-data. In order to predict which tweet would be retweeted there are two ways. One is through studying the information propagation[1] path topology to build prediction model, but it is a very difficult task to construct the topology of user networks[2]. The other way is to build prediction model based on machine learning algorithm[3].Previously, a machine learning model was built using Theil-Sen Estimator[4] and a best fit line was fit and words in a tweet were scored. We built a machine learning model which not only scores the words in a tweet but also predicts[5] the best tweet among two tweets given as the input from the user. A basic fact is that different people are interested in different kinds of tweets, and they will retweet tweets which they are interested in. First, we collect tweets of different categories from valid account of famous news media as learning corpus which acts as a dataset for our machine learning algorithm.