Urban and rural public services and management is the key problem for the development of modern society against the urban and rural resource imbalance and counter urbanization. To discover its actual contradiction, the questionnaire is the main and conventional way to survey the current situation. In this paper, we collect the three parts of Jiangsu province’s urban and rural public service data, and use the Chinese microblog comment dataset NLPCC2013 to augment the dataset. We give a two-stage framework for the text sentiment analysis of subjective answers and the global answer vector classification. The former is based on a bidirectional gated recurrent unit network for a hierarchical analysis of text sentiment from character embedding to contextualize embedding. It can output a soft label for each sub-sentence, and then locate the potential answers. According to the full scores of the subjective items, the text sentimental labels with higher probabilities are tailed with the answers of the objective items to conduct the isometric inputs of deep forest. Based on the reweighted forests’ contributions, the output layer can give a binary classification label for each input questionnaire. Experiment results demonstrate the proposed framework can predict the text sentiment accurately. Meanwhile, we analyze the source of the hot topics, and offer the policy suggestions for the future of urban-rural integration