Based on the combination of CNN and LSTM, long and short sentence network education public opinion analysis

In recent years, the importance of education has been balanced in the “national economy and people’s livelihood”. The Internet has provided a convenient way for the public to express their opinions and demands, and people’s attention to education has continued to increase. Online education public opinion and real education influence each other, and have a far-reaching impact on education. By making full use of online education public opinion, education administrators can make more scientific decisions and enhance the public’s ability to supervise education. Emotional tendency is the vane of online education public opinion, which truly reflects the opinions and attitudes of all social strata. These emotional information play a decisive role in the development of online education public opinion, so how to accurately mine the emotional semantic information in the text of online education public opinion has become a research difficulty in the field of sentiment analysis. Deep learning technology is an effective method that can automatically perform the emotional feature generation stage to mine the emotional feature information of the text. This paper takes two different texts, short text comments and long news texts generated in online education public opinion, as the practical application background, and combines deep learning technology to study the sentiment analysis methods of different length texts. Existing short text sentiment analysis methods cannot comprehensively extract the emotional features of text, and rely heavily on a large amount of language knowledge and emotional resources, so it is necessary to make full use of these unique emotional information to make the model achieve the best performance. This paper proposes a novel text sentiment analysis model that combines the advantages of convolutional neural networks and LSTMs. The model first uses the multi-head attention mechanism to learn the dependencies between words and capture emotional vocabulary in short texts. Next, LSTM is used to extract the emotional features of different granularities of the text. Then, the features are fused and the instance feature representation of the text is obtained through the global average pooling layer. Finally, judge the emotional category of the text based on the properties of the capsule. Experimental results show that the model performs well on the short text sentiment classification task and outperforms other baseline models.