Automatic Text Summarization of Chinese News Text Based on mT5 Pre-trained Model

In recent years, with the rapid development of large-scale Pre-trained models, automatic digest generation technology has been significantly improved in the field of natural language processing. In this paper, we discuss the automatic generation of Chinese news text summaries based on the mT5 Pre-trained model. mT5, as a kind of multilingual text-to-text conversion model, can effectively deal with multilingual tasks. Aiming at the characteristics of Chinese news text, a training dataset is constructed, and the mT5 model is fine-tuned to obtain a model that meets specific needs. The performance of the model in generating summaries is quantitatively analyzed using the ROUGE automatic evaluation index. The experimental results show that the mT5-based approach performs well in terms of accuracy, fluency and information coverage of news text generated summaries.