Research And Implementation On Event Logic Graph-Based Major Emergencies Summary Generation Technology

With the rapid development of big data applications,user-generated content,especially textual information,is growing exponentially,resulting in people spending a lot of time and energy on reading and processing text information.Text summarization technology based on text big data provides convenience for people to quickly extract information and effectively alleviates the problem of information overload caused by information explosion.Automatic summarization technology is widely used in the news industry,such as tracking reports of major breaking events.Major breaking events have characteristics such as a wide range of influence,fragmented information,and scattered information,which pose new challenges to the generation of summaries for major breaking events,such as difficulty in obtaining comprehensive information about the event,incomplete event descriptions,and difficulty in discovering event correlations.Based on the above challenges,the thesis proposes a technical system for generating summaries of major emergencies based on event mapping,specifically including:(1)A method based on community division and topic model for the problem that information on major emergencies is difficult to obtain comprehensively.The abrupt emergence and extensive effect of significant emergencies make timely and comprehensive information about the occurrences difficult to gather.In this research,we split large scale event logic graphs using community partitioning methods to obtain sub-graph communities of comparable events,and we use topic models to mine communities of big emergencies to achieve comprehensive access to event information.(2)To address the problem of incomplete descriptions of major emergencies,an event information complementation model based on comparative learning is proposed.When major emergencies occur,event information comes from multiple sources and presents a high degree of fragmentation,resulting in a serious lack of background knowledge of major emergencies and making it difficult to describe the completeness of the event.This paper uses a language generation model of contrast learning to generate semantically coherent and diverse summaries of major emergencies,enhancing the completeness of event descriptions.(3)An event association enhancement model based on graph structure embedding is proposed to address the problem of difficulty in discovering associations of major emergencies.The scattered information of major emergencies easily forms data silos,making event associations difficult to discover.In order to capture the rich logical associations of events in the matter graph,this paper uses a structure-aware semantic aggregation module to enhance the ability of the language model to capture event association information.The paper extracts data on major emergencies under four themes:"Earthquake","Terrorist Attack","Virus Transmission" and "Fire".The paper also shows that the proposed method of generating summaries of major emergencies is not only fast,but also effective.Based on these data,it is verified that the proposed major emergency summary generation method can not only quickly obtain comprehensive information about the event,but also reasonably add background knowledge and details of the event,and generate major emergency summaries with complete event descriptions and obvious event associations.It also integrates these methods to realize a major emergency event summary generation system based on event mapping,which improves the efficiency of major emergency event information acquisition and meets the public demand for major emergency event information.