Community detection is an important content in the field of online social networks research. The community detection algorithm based on seed expansion has the characteristics of low time complexity，high recognition accuracy，and is not restricted by the shape of the community. In recent years，it has been widely used in local community discovery of complex networks. However，this method does not consider the correlation between seeds when selecting seeds，so the number of identified community structures is large and the structure is loose. Aiming at this problem，a two-stage community discovery algorithm for multi-point seed prepartition was proposed. First，high-impact nodes in the network are identified，and high-impact nodes are aggregated using the k-means algorithm to obtain high-impact community clusters. Next，an attractiveness measurement function is proposed to selectively merge the remaining nodes in the network into the community cluster to complete the community identification task. The experimental results show that the two-stage community discovery method can find community structures with larger sizes and fewer numbers，and then capture the association between groups at the meso level.