Research on Automatic News Text Summarization Technology Based on GPT2 Model

Automatic text summarization helps human beings to read news in fragmented time, realize the news "from long to short", and get important information quickly, accurately and comprehensively. In this paper, we use GPT2 model for automatic text summarization, and rewrite GPT2LMHeadModel in the transformers package, modify the loss calculation part, and conduct 5 and 10 epoch experiments on the original news data and the split word data respectively, and use ROUGE evaluation value as the evaluation index. Rouge-1 and Rouge-2 metrics are significantly improved compared with other experiments. This model solves the inadequate and unapproachable situation existing in automatic extraction of news headlines compared with the BERT benchmark model, and achieves a better portrayal of the summary text.