Predicting the Degree of Collaboration of Researchers on Co-authorship Social Networks

Abstract

Academic social networks play a role in establishing the research collaboration by connecting researchers with each other in order to allow them sharing their professional knowledge and evaluating the performance of each individual researcher. In this paper, we mine the academic co-authorship social network in order to predict the potential degree of collaboration for each individual researcher on that network. Where predicting the degree of collaboration for each individual researcher is considered as one of the indicators for evaluating his/her potential performance. The main approach of this paper relies on using the sum of link weights specified by the total number of papers that each researcher has published with other co-authors in order to identify his/her initial degree of collaboration on the co-authorship network. This is achieved by collecting the total of link weights that each researcher has with others on the co-authorship network. Next, the obtained values of collaboration degree are used with other features extracted from the co-authorship dataset to train a supervised-learning regression model in the training phase. Consequently, the learning regression model will be able to predict potential degree of collaboration for every individual researchers on the co-authorship network in the test phase. Empirical experiments are conducted on a publicly available weighted academic co-authorship networks. The evaluation results demonstrate the effectiveness of our approach. It achieves a high performance in predicting the potential degree of collaboration of every researcher on the co-authorship network with 0.46 root mean square error (RMSE) using the linear regression model.