Collaborations among interdisciplinary scientists are playing an increasingly important role in science innovations. As it is very difficult for a researcher to master the full knowledge of his/her targeted research areas, how to find suitable collaborators of complementary expertise has turned to be a key factor for researchers to succeed. With the expansion of the Web, the availability of sheer volume of information has resulted in information overload issue and posed significant challenges on determining appropriate scientists to collaborate with effectively for research opportunities. However, current studies on collaborator recommendation ignored this phenomenon and particularly overlooked the complementarity of their expertise within a restrictive context, i.e. for a given funding proposal or a research manuscript draft. In this study we propose a complementary expertise analysis enhanced approach to retrieval experts for research collaboration. It produces recommendation list using a heuristic greedy algorithm based on probabilistic topic model, and generates experts who ought to be complemented in expertise as well as to have good ability. The proposed method has been implemented in ScholarMate research community (www.scholarmate.com). We have conducted a user study to verify the effectiveness of the proposed approach and the preliminary results show its good performance comparing to the benchmarks.