Collaborator Recommendation On Research Social Network Platforms

Collaborator seeking is an important academic activity for scholars because suitable collaborators could help improve research quality and accelerate the research process. Extensive developments in various information technologies that support scientific social platforms have resulted in the collaboration of a mass of researchers through virtual communities. Thus, developing efficient collaborator recommendation systems in these online communities is critical to promote academic collaboration.Information overload and information asymmetry are the two key issues to be addressed in the area of collaborator recommendation. In general, the potential academic collaboration contexts should be defined, and the corresponding solutions are required to provide efficient recommendations. Existing collaborator recommendation studies mainly focus on the similarity between two researchers, which is based on the expertise similarity and social network proximity. Although tremendous effort has been exerted in this area, a general framework for academic collaborator recommendation and effective recommendation mechanisms are still lacking.In this thesis, we present a general framework to formalize the collaborator recommendation issue. Two main collaborator recommendation contexts are defined, namely, the similarity-based collaborator recommendation and the collaborator recommendation in a specific context. Two efficient solutions have been proposed to address these two issues. For the similarity-based collaborator recommendation, we present a hybrid approach that integrates five heterogeneous features from the expertise relevance, social network proximity, and institutional connectivity dimensions. An expertise coverage-oriented collaborator-seeking mechanism is proposed for the collaborator recommendation in a restrictive context. A revision of the traditional latent Dirichlet allocation model is conducted to improve its performance for the corpus of documents with different effects.Two series of experiments are conducted on real-world collaboration datasets. The offline experimental results have verified and demonstrated the convincing performance of the two proposed approaches compared with those of a list of state-of-the-art benchmarks. The proposed solutions have been implemented as recommender applications on the ScholarMate platform, one of the largest academic social communities in China. The proposed methods work effectively and can guide researchers toward better knowledge sharing and collaboration.