Collaborator Recommendation Based on Multiple Information Graphs

Academic researchers work with their peers to improve the quality of their research and broaden their field of study. On the other hand, finding suitable partners for effective collaborations is challenging due to the amount of information in large-scale academic data. Different approaches for recommending partners have been put forth to address this problem. However, in order to produce high-quality recommendations, these models ignore critical content information factors such as topical relevance, affiliation, scholarly level, and publication location. Many models also ignore second-order proximity while analyzing structural information. As a result, we propose a collaborator recommendation model based on various pieces of information in order to deliver suggestions for collaborators that are more accurate. The proposed model encodes data from multiple information graphs into a low-dimensional vector space to jointly learn information from scholars. The learned scholar embeddings are then used to provide customized collaboration recommendations. Experiments on a public dataset show that our method is better than the baseline models.