The unlabeled data in multi-user social network is unbalanced, which leads to poor classification in the process of data scheduling and detection. In order to reduce the error rate of unlabeled unbalanced sample data classification in multi-user social network, a method of unlabeled unbalanced sample data classification in multi-user social network based on redundant data elimination is proposed. The subspace fusion filter detection model is used to filter redundant data and suppress anti-interference on the prior features of randomly sampled unlabeled unbalanced sample data of multi-user social network, the statistical average analysis and distributed fusion detection method of autocorrelation features are used to extract the clustering convergence feature parameters of unlabeled unbalanced sample data of multi-user social network, and the constrained evolution method of multi-level iterative regression analysis is used to estimate the classification feature quantity of unlabeled unbalanced sample data. The characteristic parameters of the classification target are input into the SVM classifier, and the adaptive weight distribution control of SVM classification is carried out in combination with the dynamic learning algorithm, which improves the adaptability of data classification and realizes the optimal classification of unlabeled unbalanced sample data in multi-user social networks. The simulation results show that this algorithm can effectively reduce the interference of redundant data when classifying unlabeled unbalanced sample data of multi-user social network, and the fidelity rate of data classification is high, and the error rate is low, which improves the dynamic data management ability of multi-user social network.