Data Mining of Scientometrics for Classifying Science Journals

Abstract

While there are several Scientometrics that can be used to assess the quality of the scientific work published in journals and conferences, yet; their validity and suitability is a great concern for stakeholders from both academia and industry. Different organizations have a different set of criteria for assessing the journals publishing scientific content. This is mostly based on the information generated from Scientometrics. A unified journal ranking system is therefore required that is acceptable to all concerned. This paper, collects data concerning Scientometrics for unified assessment of journals and proposes a mechanism of assessment using data mining methods. In order to carry out the research, big data for the proposed Scientometrics is stored in an integrated database. K Means clustering is then applied. This is to group the journals in different unsupervised clusters. The clusters are then labelled to find the exact rank of a science journal by using a state-of-the-art technique of labelling clusters. The classifier for the new instances is trained by using Naïve Bayes Classification Model. The new metrics proposed, include Eigen factor, Audience Factor, Impact Factor, Article Influence and Citations. In addition to this, the Prestige of Journal (PoJ) for the evaluation of journals is also proposed. The accuracy of both K means clustering and Naïve Bayes classification is 80%. The methods can be generalized to any problem of journal classification.