Toward the more effective identification of journals with anomalous self-citation

Abstract

Because of its important evaluative function, journal impact factors began to be manipulated by anomalous self-citations. To deal with this scientific misconduct and its undesirable influences, in this paper, an automatic classification model for journals with anomalous self-citation was constructed based on previous research. First, a training journal set and three test journal sets of normal journals and abnormal journals were established and four features were selected from a feature set. Then, a classification model was learnt using the Deep Belief Network (DBN) method, which was successfully able to identify abnormal journals in the data sets. Third, Logistic Regression and Support Vector Machine were employed to learn the classification models, the classification performances for which were then compared with the DBN model. Finally, 1138 journals in twelve subject areas from the journal Citation Report (JCR) in 2014 were chosen as empirical journal samples for the DBN model, from which 6.9 percent of empirical journals were identified as suspect journals with anomalous self-citation.