[Objective] Analyze the dynamic political sentiment polarity of American politicians’ tweets within a fixed period of time, and assist analysts in judging the direction of American politics and the future trend of China-US relations.

[Context] This research is applied in the field of think tanks or intelligence analysis. The service targets are intelligent analysts, and the data is the tweet text data sent by a specific group in a fixed time period.

[Methods]  Propose an architecture that combines multiple deep learning models and uses a dedicated tweet data set to construct a specific group to obtain an emotional polarity multi-classifier, and then introduce the time characteristics of tweets, and finally obtain the dynamic political emotional polarity of politicians .

[Result] The proposed data set of American politicians' tweets in this paper verifies the effectiveness of the proposed comprehensive architecture in this task. The accuracy of the classifier verification set reaches 80.66%, which is 8.07% higher than that of the traditional artificial neural network method. According to the emotional polarity judgments of 20 US governors and senators, the success rate is 75%.The analysis of individual dynamic political sentiment polarity can provide effective help and intelligence support for analysts.

[Conclusion] The method in this paper effectively uses a variety of deep learning techniques to assist analysts to obtain more accurate dynamic political sentiment polarity from massive Twitter text data.