Abstract
The relationship between influential tweeters and highly cited articles in the field of information sciences was analysed using Twitter data gathered by Altmetric.com from July 2011 through February 2017. The dataset consists of more than 10,000 tweets, and these mentions, retweets and followers were used to generate a connected, undirected graph. This graph reveals the most influential tweeters by identifying the largest drop in the eigenvalue of adjacency or affinity matrix of a graph when certain nodes are removed; those which, when deleted, cause the greatest drop in the eigenvalue of the graph are considered to be the most influential. The machine-learning model applied in this work utilizes a feature vector containing the accumulated sum of the rank scores of those influential users who tweet a given article, along with known altmetric features such as the user type and post counts for various social media. Finally, the supervised-learning model was trained using Random Forest and Support Vector Machine classifiers with 11 features, including the sum of the ranks of influential users who tweet a given article in our dataset. The results were analysed using Receiver Operating Characteristic (ROC) curves and Precision Recall (PR) curves, which give the commendable outcomescompared to thebaseline model. We found that, for the classification of highly cited articles, Twitter users' score for influence is the most important feature. Finally, we show that our modelwhich was trained bytaking the score for influence into considerationoutperforms the baseline, at 79% for ROC and 90% for PR with the Random Forest Model, effectively identifying the highly cited articles.