Predicting the Popularity of a Research Article on Twitter

Predicting the Popularity of a Research Article on Twitter by Ayesha Tariq (2021)

When we talk about social media Twitter is one of the most famous social media platforms. Twitter is known as a social networking service used by millions of users daily. Among different terms that are talked about on Twitter one of them is tweeting and retweeting a newly published paper. These tweets that are posted on Twitter also become part of Altmet- rics due to their (Articles’) information being recorded in Altmetrics. In this research we made predictions on if a paper will get popular on Twitter or not by using machine learning algorithms. Thirty-one Twitter and Altmetrics features were filtered to form feature space and see if based on those features a paper will get popularity on Twitter or not. To elim- inate and select features with higher importance Random Forest and Extra Trees ranking by score was used. Then backward elimination method was used on the features ranked by Random Forest method to eliminate features with lowest score. Six classifiers Random Forest, Logistic Regression, SVM, Decision Tree, Naive Bayes and AdaBoost were used respectively to check the performance of these features in classification task. This feature space included features that are initially available at the time of publication or Twitter post. Our experimental results showed that by using top ten features ranked by Random Forest and top eleven features ranked by Extra Trees two classifiers Random Forest and Decision Tree performed best by giving higher accuracies.