Text sentiment polarity classification based on TextCNN-SVM combination model

In the face of a large amount of information, managing, filtering, and screening information has become effectively one of the focuses in information science, especially in the accurate processing of text information. In this paper, a TextCNN-SVM combination model for text sentiment polarity classification was proposed and implemented in the TensorFlow framework. Combined with the ability of TextCNN to extract text eigenvalues well and the classification function of Support Vector Machine (SVM) to find the maximum segmentation hyperplane, we improve the shortcoming of the SoftMax layer, which is weaker than the machine learning algorithm in terms of classification ability and generalization ability. Compared with the traditional TextCNN model, the accuracy of the combined model has been improved by about 12.37%.