Resrarch On Single Document Summarization Metnod Based On Graph Model Combined With Topological Potiencial

With the development of the internet,the amount of text data we are exposed to every day is also exploding,and massive data increases the difficulty of obtaining critical information.Automated text summarization can alleviate this challenge,optimizing the quality and improving the accuracy of text summaries has become a research hotspot in the field of natural language processing.Researchers are striving to obtain key information from a large amount of text data through various methods,saving readers’ time and energy.Currently,text summarization based on Text Rank algorithm only considers the similarity between nodes,neglecting the importance of sentences and other information,which affects the quality of extracting text abstracts.Based on existing research,this paper proposes a graph model single document summarization method combining topological potential.This method comprehensively considers sentence importance and similarity,and optimizes the edge weight relationship of the graph model in the Text Rank algorithm.The main work of this article is to fuse topological potential and similarity,measure the importance of sentences using topological potential,and compare the quality of extracted text abstracts before and after the introduction of topological potential.Firstly,Bert model is selected for vector representation of text information.Bert model uses the method of dynamically extracting vector,which can obtain more in-depth semantic information based on context,making the obtained text vector representation more flexible.Secondly,the idea of topological potential is introduced,and two similarity measures,cosine similarity and word mover’s distance,are combined with topological potential to calculate topological potential values.The importance of sentences is measured by the size of topological potential values.The similarity and importance of sentences are comprehensively considered to further optimize the edge weight structure.Finally,the Text Rank algorithm is used to calculate the sentence score and sort the sentences,and the top ranked sentences are output as summaries.The experiment was conducted on two common data sets,TTNews and CNews,and the results showed that the quality of abstracts generated using this method was significantly improved in terms of the three evaluation indicators: route-1,route-2,and route-L.