Research On Abstractive Summarization Technology Of Public Opinion News Based On Element Extraction

According to the data of China Internet Network Information Center(CNNIC),as of March 2020,the number of Internet users in China was 904 million,with an Internet penetration rate of 64.5%.With the continuous expansion of Internet users,the rate of dissemination of Internet public opinion news is getting faster and wider.In particular,in early 2020,the level of attention to public opinion news,affected by the new coronary pneumonia epidemic,further showed a rapid growth.In addition,as major online platforms seize the Internet news market,the same hotspot of public opinion is often accompanied by a variety of complicated and highly redundant public opinion information.It takes a lot of time and effort for netizens to quickly and comprehensively understand the main information of a hotspot and follow its next development.Therefore,in order to improve the reading efficiency of netizens and to assist them in following the dynamic changes of public opinion hotspots,the following quantitative work has been carried out in this topic.Firstly,this thesis studies the element extraction techniques for public opinion news with the aim of extracting structured data from a large number of unstructured news texts on the Internet,as external knowledge.This thesis defines the element extraction which is a non-traditional natural language processing task,clarifies the granularity hierarchy of element extraction,and elucidates the differences and similarities between different granular elements.Then we use different levels of natural language processing techniques to implement them.For the argument-level element extraction study,this thesis employs semantic role labeling techniques.For the entity-level element extraction study,this thesis employs open information extraction techniques.Secondly,this thesis studies the abstractive summarization technology for public opinion news.Its purpose is to use automatic summarization algorithm to analyze and summarize the long-length public opinion news text and output a concise and concise text summary.By analyzing the problems of the existing methods,this topic proposes a solution to the knowledge-driven generative summarization model based on global coding information,and attempts to combine the feature extraction results with automatic summarization technology to improve the final performance of the summarization systemThe technology proposed in this thesis has achieved good performance in various evaluation metrics of different corpus and has been applied in the actual public opinion monitoring system related to the thesis.