With the arrival of the big data era, the amount of micro-blog users and texts is constantly increasing, and research on personalized recommendation algorithm for micro-blog texts is becoming more and more urgent. In consideration of the impact of user’s interests, trust transfer, time factor and social network, we proposed a new method for personalized recommendation. The method is based on community discovery, and recommends personalized micro-blog texts for users with the improved user model, which can use the social network of micro-blog platform effectively and optimize the utility function for micro-blog recommendation. Firstly, we used a multidimensional vector to represent the stereoscopic user model. Secondly, we proposed the improved k-means algorithm to extract the local community of users, which was also used to get the recommend micro-blog texts. Finally, the top-n micro-blog contents sorted by the effect function were recommended. We used a large number of real data to verify the algorithm proposed in this paper, and compared our method with some existing algorithms.