In this paper, the variational self-encoder is dissected, and the depth-embedded variational self-clustering model based on it is constructed, and the loss function and optimization function are designed. By enhancing the information of textual semantic representation, the semantic enhancement module is constructed based on the deep variational inference algorithm, the network structure of the semantic enhancement module is designed, and the SEVAE model of textual clustering based on the semantic embedding of DataSmart is constructed by combining textual clustering layer and dataSmart connotation. On this basis, the comparative experiments of the SEVAE model are constructed to analyze the convergence and clustering performance of the constructed model, explore the connotative features of Data Sensemaking, and analyze the effect of Data Sensemaking through empirical evidence. The results show that the clustering assignment process converges at the 20th iteration round, which significantly improves the feature representation, divides the 8483 texts into 3 groups, and the Data Thinking has the characteristics of Data Thinking, Data Information, and Data Quantification, with a positive, positive effect of P<0.1 for all the control indicators. Data Thinking has the power to promote patriotic sentiment, patriotism, and social cognition.