Research And Application Of Automatic Text Summary Technology Based On Graph Neural Network

The advent of the digital age makes people’s access to information more efficient and convenient,but also makes a large amount of information filled in people’s lives.In order to meet people’s need to quickly obtain information from massive information,the automatic text summary technology has entered people’s vision.Automatic text summary technology is the use of computer technology,according to the needs of users to extract the most important information content in the long text,after a summary,finally got a simplified short text process,which can effectively improve the efficiency of users browsing information.Graph neural network is a kind of neural network that can learn feature rules from graph structure data.It can realize the processing of non-Euclidean spatial data by using the complex network structure,which is beneficial to capture the relationship between sentences in the abstracted text summary.In this study,the automatic text summary algorithm of graph neural network HSSG,based on the binary graph built by word nodes and sentence nodes,superposition between sentence nodes,and introduces the spatial logical relationship between sentence nodes,making the generated sentence node features not only contains the word information,but also contains the information of other sentence nodes.Therefore,the characteristics of the generated sentence nodes are more rich,and the text summary is more accurate.Moreover,considering that emotion,as an important psychological phenomenon of human beings,exists in the text formed by human language,it directly affects the importance of the selected sentences in the text summary.Therefore,this study introduced emotion features based on the HSSG model and constructed the HSSEG model,which realized the extraction of the text summary of emotion perception.Based on the proposed HSSG model,this study further constructs the automatic text summary model of graph neural network of multi-type nodes HSSMG,build the semantic nodes,and generate binary graph between semantic nodes and sentence nodes,add to the proposed HSSG model structure to make the final generated sentence node information more comprehensive.Based on this,emotion features are introduced to build a multi-type node model of emotion perception,HSSMEG,which achieves better performance and generates more accurate summaries.A specific application of the HSSMEG-based model,which is applied to the text summary generation of an audio file.The application can help users to quickly obtain key information in the massive voice information,and provide it to users in the form of text.