Research On The Technology Of Emergency Detection Based On Graph Network

Due to the higher degree of liberalization of social media compared to traditional media,emergency information on social media is easy to spread widely and be maliciously spread and tampered with by intentional individuals,achieving its hidden purpose,posing great risks to the security of the network environment and the long-term stability of society.Therefore,it is of great significance to quickly and accurately detect emergency related information from a massive amount of social media information,analyze the emotional state of the public towards the event based on the detected emergency,and take relevant measures as soon as possible to maintain network security and social stability.This thesis has done the following work:(1)Propose a heterogeneous information network based on time division to organize data.The traditional event detection methods do not analyze the time factor.This thesis uses a multi-dimensional time window to separate the tweet publishing time and designs a time based weight calculation method to further improve the accuracy of event detection.(2)Add influencing factors to the event detection model.Traditional event detection methods assign the same weight to all tweets,but in social media environments,there are differences between users,such as differences in influence between official institutional users and an independent user.Secondly,there are differences between tweets,which are mainly reflected in the number of reposts,likes,and comments in the tweets.This thesis designs an impact calculation method by examining the relevant data of the tweet publisher and the tweet itself,and integrates the results of the impact calculation into the event detection model,improving the accuracy and efficiency of event detection.(3)Propose an aggregated event sentiment analysis model.Traditional models often rely solely on text vectors or emotional words for emotional analysis of events.This thesis proposes a joint model method based on the results of event detection,which independently calculates the emotional polarity of all tweets in the set of tweets to which the event belongs,and ultimately aggregates them into the emotional polarity results of the event.This method improves the accuracy and efficiency of event sentiment analysis.Finally,this thesis designed experiments to verify the performance of each module in practical work.The model was validated using a supplementary dataset based on the public dataset which named Crisis Lex.And on this basis,a set of event detection and analysis system based on social media data was developed...