Research On Science Citation Network Analysis And Its Applications

Citation network analysis attracts great attention for its huge value of research and application. However, most of the research work concerned only static network analysis, lacking of empirical research and prediction of dynamic development of papers’ citation trend. This paper studies the dynamic characteristics of citation network development in detail and further proposes a number of algorithms to predict the trends of citation. The main work is as follows:1. We proposed a kind of algorithm to forecast papers’ potential value based on citing diversity. There exist some limitations for current methods evaluating science paper according to the number of citation directly, ignoring the fact that citation is developing dynamically. Through empirical analysis on real data sets, we found that filed diversity and time diversity of citing can reflect the potential value of being cited continuously in future to a certain extent, thus we modify the exist algorithm of papers’ potential value to propose another algorithm based on citing diversity. Experimental results show that compared with the original algorithm, this method can tap the potential value of the paper and forecast more precisely and efficiently in the absence of author information.2. We proposed a method to sort science papers dynamically based on sustained attention decay. Existing algorithms for papers’ dynamic ranking have problems such as low accuracy or time consuming. Through empirical analysis of real data sets, we find that there is a strong correlation between future citations and annual citation of papers. Thus we put forward the concept of papers’ sustained attention. Further, combined with time decay effect in citation network, we design an algorithm to rank paper in dynamic. The calculation of this method is relatively simple, especially applicable for ranking papers quickly in the mass literature. Experimental results on two typical data sets show that the proposed algorithm improves the prediction accuracy of 30% compared with the existing algorithms.3. We proposed a prediction algorithm which can present the specific number of papers’ future citation precisely. Existing prediction methods of citation network mostly can only evaluate papers’ future ranking and classification. They are unable to carry out the specific numerical predictions. In this paper, using papers published in the same period as the research object, we analyze two factors- the growth of new references and paper’s own quality in peel, which heavily impact paper’s citation trend. And then we put forward an algorithm to predict the future citation number of individual paper. Extensive experiments demonstrate that the proposed algorithm can not only forecast the future popularity of individual papers accurately but also have the ability of long-term predicting for papers’ future rank in the system.