Research And Implementation Of Text Summarization Based On Key Information Discovery

With the development of Internet and the popularity of smart devices,digital media such as Weibo and Toutiao is becoming widely popular.The amount of online information grows rapidly and information overload has become increasingly serious.Faced with massive information,people need to spend a lot of time reading and analyzing.Therefore,how to quickly obtain effective information has become a common concern of all sectors of society,and automatic text summarization technology is the core to solve this problem.Previous methods can be divided into two categories:ive and extractive.Compared with extractive methods,the ive strategies use advanced content understanding and text generation models,improving the overall logicality and fluency of generated summaries.However,existing abstractive approaches still have the problems of incomplete text modeling,weak content distinguishing ability,and lack of long-term dependencies,which leads to repeated sentences,redundant secondary information,and incomplete information points.To address these issues,this paper proposes a convolutional based encoder-decoder abstractive framework GCTTS and applies the model on the news reports.Our embedding method has three novel extensions.Specifically,(1)we design a graph-based text representation algorithm.The algorithm starts from the article structure,word semantics,and topic relevance,while taking into account the local and global features,and finally improves the text modeling effectively;(2)we design content-selection network based on graph convolution.Through the effective learning of scattered information points and rewriting mechanism,the accuracy of the model information selection is improved;(3)Compared with recurrent neural networks,the multi-layer convolutional architectures are better at capturing long-range dependencies.Experimental results demonstrate that our model has been improved on the ROUGE metrics by nearly two percent and generated summaries are more brief and complete.In terms of technology landing,we have developed a complete news analysis system with automatic data acquisition,reports clustering and summaries display.