With the development of Web2.0, Internet users can use virtual social network platforms to interact with each other freely and contribute their individual brilliance voluntarily. In the collaborative creation process, common-based innovation network gradually formed, featured by reciprocity and coopetition relationship. Collaborative innovation relies on the wisdom of crowds, they share they ideas together and further create new knowledge. As a result, knowledge flow becomes more flexibly and knowledge outflow is strengthened greatly in the common-base innovation network. This paper attempts to investigate (1) how direction of knowledge flow (inflow or outflow) influence the performance of new product development, and (2) the two-side effect of social capital based on the analysis of knowledge flow characteristics (homogeneous or heterogeneous) which is determined by social network structure. Using the data from Sourceforge.net, we construct the common-based innovation network based on membership information and identify the direction of knowledge flow based on role information. Results show innovation teams who receive rather than provide knowledge can achieve better innovation performance. Besides, degree centrality, which is featured by homogeneous interaction, can improve innovation performance in the first place, but it will reduce innovation performance after degree centrality being too large. Betweenness centrality, which is featured by heterogeneous interaction, has positive effect on innovation performance.