Abstract: Friend recommendation is an effective method for establishing an online community. However, over frequent recommendations may be the opposite and become nuisances to users. To improve users' experience, a new method of friend recommendation is proposed via many-to-many assignment. This method limits the number of recommended and accepted friends. It takes as the application background the website http://www.scholat.com/, which is a large higher education and research collaboration platform. Recommendation is modeled via Role-Based Collaboration and its E-CARGO model. After that, the Kuhn-Munkres with Backtracking (KM_B) algorithm is used to solve the optimal assignment of the proposed method. Simulation experiments show that the proposed recommendation method is friendly, efficient and accurate. It can improve the online community recommendation mechanisms, which can support the development of a virtual society.