Most of the previous studies on scientific collaborator recommendation are based on social proximity analysis to suggest collaborators. However, the extracted homogeneous features cannot well represent the multiple factors which may implicitly affect the future scientific collaboration. In this paper we propose an approach based on the multiple heterogeneous network features, which has produced good results in our experiments based on a dataset of more than 30,000 ISI papers. This method can help solving the similar problems of people to people recommendation. It generates high quality expert's profiles via integrating research expertise, co-author network characteristics and researchers' institutional connectivity (local and global) through a SVM-Rank based information merging mechanism to perform intelligent matching. The generated comprehensive profiles alleviate information asymmetry and the multiple similarity measures overcome problems related to information overloading. The proposed method has been implemented in ScholarMate research network (www.scholarmate.com) which is a research 2.0 innovation, promoting research collaboration in virtual scientific community.
展开▼
