Mining opinion leaders in social network is important for analysis of information dissemination and evolution of public opinion.This paper conducts the study on this problem considering structural features

behavior and emotional characteristics comprehensively.Firstly

we extract topics from micro-blogging texts

and get user communities according to the topic division

and an interactive network topology of topic community is built with the following relationships.Then

three kinds of user feature are gained from different aspect:network structure

user behavior and user sentiment.Finally

according to the analysis of users' influence distribution

opinion leaders mining algorithm MFP (Multi-Feature PageRank) is proposed.Experiments show that the algorithm can obtain the potential opinion leader nodes effectively

and have a good performance in support rate from other user nodes.