Abstarct: In the last decade, the high penetration rate of virtual social networks among individuals in communities and the frequency and availability of their data has attracted the attention of many researchers to the analysis of social networks. One of the important applications of virtual social network analysis is event detection and prediction. Existing event prediction methods focus only on one specific event, while what is addressed in this study is the prediction of all predictable events. In this research, the event refers to something that causes significant changes in some parameters and important features of the virtual social network. A feature that has been considered in the proposed method is changes in the rate of sending messages to the social network. In the proposed method, baxsed on a fixed time window, the operation of splitting and cleaning tweets is performed. The tweets are then categorized using the Non-negative Matrix Factorization (NMF), a matrix analysis method, and incremental clustering distance dependent Chinese Restaurant Process (ddCRP). After that, the occurrence of the event is predicted baxsed on the changes in the rate of tweets entering each cluster. Finally, a descxription of the event is displayed in a few repetitive words in each cluster. The proposed method, baxsed on a dataset of Twitter, contains approximately 12 million tweets covering 1940 events, is evaluated, and succeeds in predicting future events with 85% accuracy.