News text summary generation is an important part of the news production process, which is based on the news text keywords to highly condense the text content and help the news information to be quickly understood and smoothly digested. This paper combines the Htmlparser tool with an improved template-based SST tree method to obtain key news information text in web pages. Aiming at the characteristics of large data volume and more isolated points of news event information, the data metaclustering algorithm under the idea of density clustering is proposed for feature mining of news information text. On this basis, the LDA topic model is used to obtain the topic difference influence degree of information text keywords, the fusion of LSTM model and word2ec model is used to calculate the semantic relevance influence degree, and the two influence degree features are put into the TextRank algorithm to calculate the keyword importance. A generative code summarization model structured by multi-component encoder, encoder and keyword guidance is formed, and the keyword information is used to guide the encoder to generate summaries corresponding to the topics of the original text, so as to transform structured text data into short code summaries. The keyword-guided news generative summarization model generates new word rates between 18.97% and 91.01% in different structural tasks, and the three evaluation indexes of ROUGE-1, ROUGE-2, and ROUGE-L are as high as 46.36%, 29.78%, and 41.49%, respectively. The automatic generation of abstracts with artificial intelligence supported by big data technology realizes the information summarization and framework refinement of news text, and promotes the double improvement of news production quality and efficiency.
