Multi-label Classification of News Topics Based on Universal Sentence Encoder

News topic classification is a topic division of news based on the title or content of the news. Correct news topic classification can improve the efficiency of information retrieval and optimize the content recommendation system. Encoding the text information into the computer for classification requires coding, and good coding can improve the accuracy of classification. And in order to ensure the timeliness of news, faster classification speed is needed. This paper studies the multi-label classification method of news topics based on universal sentence encoder. The data set used is MN-DS. Firstly, the title and content of the news are encoded by a universal sentence encoder, and then input into different classifiers (Multinomial NB, Logistic Regression, SVC Classifier) for classification, and achieved very good results. Finally, it is compared with other coding methods (TF-IDF, Glove). The experimental results show that the universal sentence encoder has the highest classification F1 value on the title, reaching 0.803, indicating that the universal sentence encoder is more suitable for the coding of short sentences. Moreover, the speed of universal sentence encoder is much higher than other coding methods.