Automatic Classification of Government Texts Based on Improved CNN and Skip-gram Models

Aiming at the characteristics of government texts with rich semantic information, the impact of convolutional neural network models on the classification and recognition of government text features is studied, and an optimization method to assist in the classification of official government information resources is proposed so that government staff can better investigate public opinion. Based on machine learning and deep learning theories, this paper conducts an experimental study on automatic classification and recognition of texts using convolutional neural network models and conducts a comprehensive comparative analysis from the perspective of information details and core topic sentences. After data cleaning, Word2Vec is used to construct a word vector matrix, and this information is integrated into the convolutional neural network model for comparison and optimization. And based on the use of a convolutional neural network based on the improved convolutional layer technology, combined with a Skip-gram word vector model, the text is converted into low-latitude feature vectors as the input of the improved convolutional neural network model to achieve the automatic classification of government texts. The experimental results show that the model of this study achieves 92.7% F1 mean and 93.4% ACC in the evaluation of automatic classification of government texts, which not only solves the problem of excessive dimensionality but also significantly improves the execution performance and classification accuracy compared with the traditional methods, and effectively solves the automatic classification of information contents and labels.