Temporal citation network-based feature extraction for cited count prediction

Abstract

The academic data that keeps the advanced knowledge of mankind continues to increase. Accordingly, researchers have been actively conducted research to find important ones among the academic data. This study presents new features for citation count prediction problem. The new features are derived from the network centrality analysis with time transition variance and are compared with the existing author, venue, and content features to verify their excellence. We use coefficient of determination (R2) as a performance measure, and it has been confirmed that our proposed features are more useful for predicting the citation count than the existing features. Along with presenting new features, we have also attempted time-series analysis to observe whether the features used in the prediction change their influence with time. Thus, we have found that the influence of features does not change much over time.