Trend-Based Citation Count Prediction for Research Articles

摘要

This paper aims to predict the future impact, measured by the citation count, of any papers of interest. While existing studies utilized the features related to the paper content or publication information to do Citation Count Prediction (CCP), we propose to leverage the citation count trend of a paper and develop a Trend-based Citation Count Prediction (T-CCP) model. By observing the citation count fluctuation of a paper along with time, we identify five typical citation trends: early burst, middle burst, late burst, multi bursts, and no bursts. T-CCP first performs Citation Trend Classification (CTC) to detect the citation trend of a paper, and then learns the predictive function for each trend to predict the citation count. We investigate two categories of features for CCP, CTC, and T-CCP: the publication features, including author, venue, expertise, social, and reinforcement features, and the early citation behaviors, including citation statistical and structural features. Experiments conducted on the Arnet-Miner citation dataset exhibit promising results that T-CCP outperforms CCP and the proposed features are more effective than conventional ones.