The result of the research is a proposed model for text analysis and identifying the subject and content of texts on Twitter. In this model, two main phases are implemented for classification. In text mining problems and in text mining tasks in general, because the data used is unstructured text, there is a preprocessing phase to extract the feature from this unstructured data. Done. In the second phase of the proposed method, a multilayer neural network algorithm and random graphs are used to classify the texts. In fact, this algorithm is a method for classifying a text based on the training model. The results show a significant improvement. Comparing the proposed method with other methods, according to the results, we found that the proposed algorithm has a high percentage of improvement in accuracy and has a better performance than other methods. All the presented statistics and simulation output results of the proposed method are based on the implementation in MATLAB software.