Research On Text Summary Generation Method Based On Fact Fusio

With the continuous innovation of deep learning technology,text summarization has become one of the important research directions in the field of natural language processing.In various industries such as financial analysis,medical data collation,and legal fields,text summarization has become an indispensable auxiliary decision-making tool.With the advancement of technology,generative summarization has become a current research hotspot.The advantage of generative summarization is that it can more flexibly express the content in the original text,but the quality and accuracy of the generated results depend on the quality and accuracy of the summarization model.Therefore,there are still certain challenges and difficulties in the research and application of generative expressions in the field of natural language processing.Generative text summarization models are highly flexible.Despite the continuous development of technology,there are still some problems in the generative text summarization model,for example: the generative summarization model may have factual errors,which will seriously affect the usability of the summary;at the same time,the model often lacks the guidance of the summary template when generating the The content of the text cannot be completely faithful to the meaning of the original text.Therefore,this paper aims to explore how to fuse the factual information of the text to generate more accurate and complete summaries.The main research contents are as follows:This paper proposes a model framework for text summarization based on the fusion of factual relationship and keyword information.The model framework uses the public information extraction technology to extract the factual relationship existing in the input text,and automatically extracts the keyword information in the text through the sequence annotation model.Finally,these two pieces of information are fused to generate accurate and comprehensive text summaries.Experiments show that this method has improved the summaries generated by CNN/Daily Mail and XSum datasets compared with other baseline models,and successfully alleviated the problem of factual errors in summaries.This paper presents a method for guiding text summarization based on sentence templates.In the task of text summarization,sentences are an important part of summarization,and many studies focus on the selection of key sentences in text.Inspired by this,this study proposes a method that uses key sentences in the text as summary templates to guide the generation of text summaries.The abstract template contains a lot of factual information in the original text,select appropriate sentences to form the abstract template,and send it to the model code.Multiple sets of experiments show that,compared with other models,the summary generated by this method on the dataset CNN/Daily Mail is more faithful to the original text,and the proportion of out-of-set words is lower.