BMCR: A collaborator recommendation model based on Scholars’ academic level

Abstract

Collaborator recommendation helps scholars find suitable collaborators, which is useful for the development of scientific research and the transformation of scientific and technological achievements. However, the gap between scholars’ academic level seriously affects the establishment of cooperative relations. In this paper, a new model named Best Matching Collaborator Recommendation (BMCR) is proposed. BMCR model looks for the best matching collaborators for scholars from a new perspective of academic level. First, we define the academic level index of scholars and cluster scholars with this index. Next, the research topics of scholars are extracted from their published papers by using the topic extraction model. Then, for each cluster, a scholar-topic graph is built with the scholars and their topics. Finally, given the target scholar, the biased random walk with restart algorithm is carried out on the scholar-topic graph to find potential collaborators. Experiments on the MAG, DBLP, C-DBLP datasets show that our model improves the feasibility of cooperation between the target scholars and recommended scholars. Compared with other methods, our model has improved precision rate, recall rate and F1 score by 5.3%, 2.5%, and 4% respectively. The experimental results indicate that scholars’ academic level is a key reference factor in the collaborator recommendation, and that scholars with similar academic level are more likely to achieve practical cooperation. The findings are not only of great significance in recommending collaborators to scholars, but also can be extended to other fields to solve the problem of collaborator recommendation.