Sentiment Analysis of Comment Texts on Converged Media Platforms based on BERT Model

Abstract

The development of mobile Internet, smart terminals and cloud computing has promoted the advent of the era of media convergence. And Converged Media Centers in various regions had gradually been established, assuming the responsibility for disseminating news and current affairs and guiding and supervising public opinion. The study of comment texts under the converged media platforms is more targeted. In addition, this research is of great significance to the supervision and guidance of local public. This paper applies the BERT (Bidirectional Encoder Representation from Transformers) model to the sentiment analysis of comment texts on a converged media platform. In this experiment, the data used are the comment texts collected by our laboratory under the videos, which were posted by a part of the Converged Media Center accounts on the Douyin platform. By fine-tuning the pre-trained BERT model, and then the results obtained are classified and processed through the constructed neural network model. The accuracy of the validation set is 85.83%. Compared with other models, this method has a significant improvement in classification accuracy. When it is applied to a converged media platform, it is very possible to obtain correct classification results.