In the era of big data, innovative marketing of internet finance confronts with new challenges. How to provide users with precise customized financial service, has become a problem of further expansion of internet financial market. Based on this background, the research integrates the ideology of concept hierarchy into the traditional personalized recommendation algorithm, and proposes an improved trick, which is punishing user similarity computation of popular items. And the research integrates the ideology of community mining into recommendation algorithm, making EO algorithm as the basis of community division algorithm, which combines improved index of user similarity and the Q of community discovery to build a new model of community discovery and divide the structure of social network. Optimizing the procedure of recommendation system solves the problems of data sparsity, cold start and system scalability. In the end, selecting the data from the Mint, which is a financial institution in America, demonstrates the improved effect that the two innovative schemes proposed in this research contribute to the accuracy of the recommendation system and the precision of community discovery.