Abstract:In order to analyze the impact of social networks on the spread of public opinion

the multi-source network community structure characteristics and its propagation characteristics was studied. Furthermore

the overlapping community was discussed using COPRA algorithm and LFM algorithm in social network and the improved algorithm of LFM based on filtering node degree——NF-LFM algorithm was proposed. In this algorithm

the node whose node degree is less than a certain threshold in the friend relation network was filtered

and then the rest of the network was divided into community structure. The results showed that: 1) The social networks such as Renren

Qzone and Microblog have obvious characteristics of the community structure

of which Renren and Qzone is stronger than Microblog. 2) Without taking into account the user activity

the diffusion range of public opinion information in Renren is larger than the others. The proposed method can not only solve the resolution problem of the existing algorithm in the community division results effectively

but also make up for the shortcomings of the LFM algorithm

which is caught in an infinite iterative process and leads to high time complexity. It is also in line with expected results when is applied to the classical data set. The results of this paper can help researchers to understand the impact on public opinion spread in community. It also has important significance for the discovery of network group incidents and guidance of public sentiment.