Recently, there are some problems like extracting hardly and lacking classification methods in stylistic classification of social networks. Combining network stylistic diversity, multi-attribution and dynamic characteristics.A attribute fusion and thesaurus associated method based multi-agent has been proposed from feature extraction. Firstly, it extracts the basic attributes of keywords and meaning of characteristics. Then, a multi-agent fusion classification model has been established with the interaction of multi-agent and it also gives the algorithm of the model. The experimental results show that this method which compares with the traditional single fusion classification classifier and other multi-classifier fusion classification not only achieves the high-precision network stylistic classification in semantic network through Semantic features extraction but also receives Social Network stylistic classification's automation. The method has a higher accuracy classification and stability.