A community partition al gorithm is designed based on theme and connection. Both theme and connection sim ilarity of nodes are integrated in the algorithm, which also adopts a localized way to avoid the searching of good initial nodes. In the proposed algorithm, loc al modularity is accepted as a terminating condition of community partition. The algorithm is applied to a set of Twitter users who had joined into the topics r elated to Haiti earthquake. Three baseline community partition algorithms, i.e., an algorithm simply based on link, an algorithm simply based on topic, and an a lgorithm based on both topic and link, are also applied to the same data set. Experiment results show that the proposed algorithm is [JP2]more advantageous than the three baseline algorithms according to the measurement of purity and entropy.