Co-authorship prediction method based on degree of gravity and article keywords similarity

Abstract

Link prediction is a technique for predicting future relationships among candidate node pairs. The co-authorship prediction measures the candidate by examining the unobserved node pairs using the link prediction technique. Previous studies have proposed co-authorship prediction and focused solely on using a topology or content articles to conduct the co-authorship prediction. However, many unobserved node pairs hinder the co-authorship prediction process. A new co-authorship prediction method is required by considering both topological information and research interest due to the authors collaborating to publish scientific papers based on research similarities, although still considering the network topology. The objective of this research is to propose a co-authorship prediction method based on a two-phase process: pruning candidate node pairs based on article content similarities to avoid a large number of candidate co-authors and predicting potential co-authors based on the Degree of Gravity for Link Prediction (DGLP) method. The proposed method is examined using the real-world co-authorship network and assessed using the area under the curve and the paired samples t-test to show a significant improvement. The experiment results show that combining DGLP, keyword extraction, and keyword similarities can help obtain the best performance and outperform the benchmark methods for co-authorship prediction in the unweighted network.