Network ideological and political education is a great extension of school education, but the complex and confusing network information is also easy to give students bad emotional guidance. In this paper, microblogging data and topic clustering technology are used to obtain the emotional text data of network ideological and political education, based on the LDA model for topic feature extraction, to remove the data that has a lower degree of relevance to ideological and political education emotion in the data. Then, combined with the Jieba segmentation tool, based on the TF-IDF algorithm, for keyword extraction of the data that has been segmented. The text semantic features extracted based on the Word2Vec word vector model are used as the input information of this model, and the unique memory unit and gating structure of LSTM are utilized to solve the problem of storing memory of a priori knowledge and the problem of gradient dissipation caused by too long time threshold in long text classification. The Word2Vec-LSTM text sentiment classification model is developed and trained using Python and related machine learning libraries, and comparative experiments are carried out to validate the algorithm’s effectiveness and feasibility. The performance of Word2Vec-LSTM sentiment analysis on the dataset is higher than Word2Vec and LSTM algorithms. In addition, the value of sentiment perplexity is the smallest when the number of topics is set to be about 8, and its value is 0.542, and the positive sentiment density reaches the maximum value (0.648) as the sentiment score grows to 0.48. This paper explores the application of emotion analysis results in network ideological and political education to further improve the construction of a network ideological and political education system.