Which can better predict the future success of articles? Bibliometric indices or alternative metrics

Abstract

In this paper, we made a survey on the prediction capability of bibliometric indices and alternative metrics on the future success of articles by establishing a machine learning framework. Twenty-three bibliometric and alternative indices were collected to establish the feature space for the predication task. In order to eliminate the possible redundancy in feature space, three feature selection techniques of Relief-F, principal component analysis and entropy weighted method were used to rank the features according to their contribution to the original data set. Combining the fractal dimension of the data set, the intrinsic features which can better represent the original feature space were extracted. Three classifiers of Naive Bayes, KNN and random forest were performed to detect the classification performance of these features. Experimental results show that both bibliometric indices and alternative metrics are beneficial to articles’ growth. Early citation features, early Web usage statistics, as well as the reputation of the first author are the most valuable indicators in making an article more influential in the future.