(School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China)
Abstract： Social media account classification methods start from the attribute information of the account, construct the account features and classify the account, which is very important for mining valuable information from the massive social media data. Existing social media account classification is generally based on extracting features from the information posted by users, which has the problems of incomplete description of account information and low effectiveness of classification. To solve the problems above, the paper proposes a social media account classification method based on multimodal feature fusion. The method uses tensor analysis to fuse the multimodal features expressed by the account after comprehensively considering the information of the account′s own attributes, the text, and the social relationships between the accounts. Compared with the existing methods, the method proposed in this paper can better utilize the various information of accounts and obtain better classification results. Through experiments, the method in this paper achieves an accuracy rate of 93.74%.
Key words : social media; account classification; feature fusion; tensor decomposition
