Research On Social Media User Latent Attribute Recognition

Nowadays,the popularity of social media has accumulated a huge and valuable user data resource.These massive heterogeneous data have brought new breakthroughs to industry and academia in the field of large data mining,and the identity recognition task of social media users has emerged as the times require.Social media users identify latent attributes,and use existing scale data to automatically infer unknown attributes and potential characteristics of user communities.Based on the different user data of micro-blog and Zhihu,two ways are put forward:1.Latent attribute recognition method for social media users using Labeled LDA modelIn the paper,we use the Labeled LDA model to identify the latent attribute of the Zhihu community,the social orientation of the social media-career orientation.Compared with micro-blog user text data,the Zhihu community user text is more than thousands of long text data and is large and needs complex data cleaning to make use of it.In the paper,we try to use user behavior characteristics--fans' users' latent attributes to "opinion leaders" users' attention behavior--recognition of career orientation.Due to the existence of normative texts in various industries in the Zhihu community,we directly uses the existing standard text as tags,and uses Labeled LDA model to identify latent attribute.Using user behavior characteristics to identify latent attributes,the performance of career orientation recognition is 5 percentage points higher than that of text statistics.2.Latent attribute recognition method for social media users based on recurrent neural networksBased on recursive neural network,this method integrates multiple features to identify the latent attribute of social media users-gender attributes.The performance of the traditional SVM model depends on the selection characteristics.When the feature selection is bad,the classification performance of the SVM model is not good.In addition,when the traditional feature data is too large,it is easy to cause data sparseness and reduce the performance of the system.This paper analyzes social media Sina micro-blog,selects user tags and user attention content as fusion features,and identifies user latent attributes based on recurrent neural network.In addition,the traditional SVM model and dictionary method were used as the control group.The experimental results show that the system performance is better when the recurrent neural network is based on the experiment of the latent attribute recognition task based on the characteristics of the traditional features and the features of the concerned content.In the above two experiments,we find that the performance of the recognition system is related to the user scale.The higher the user scale is,the better the performance of the recognition system is.