Research On Network Hot Event Prediction Algorithm Based On Information Fusion

With the vigorous development of Internet technology,social networks have penetrated into all aspects of society and public life.The dissemination of divergent hot events through social networks can,to a certain extent,reflect current social issues and popular tendencies.However,hot social events are prone to negative comments or rumors,resulting in certain negative impacts and pressure on public opinion.Therefore,the research on prediction of hot events has very important practical significance,which can help relevant departments to monitor public opinion,do a good job in emergency management,and maintain public safety.Currently,in hot event prediction tasks,there is a lack of semantic information at the beginning of an event,and information is diverse,sparse,and dynamically changing during the propagation process.As a result,existing methods have problems such as insufficient utilization of initial event information,and insufficient mining of dynamic information features,which affect the accuracy of model prediction.Based on the above problems,this thesis focuses on the network hot event prediction algorithm based on information fusion,and the specific research contents are as follows:1.For the task of early event prediction,an early event prediction model based on Bayesian network is proposed.The existing early prediction algorithms have the problems of incomplete utilization of feature information and high complexity of feature extraction.For this reason,this thesis designs and extracts emotional characteristics,forwarding frequency characteristics,user characteristics and network structure characteristics,constructs Bayesian network model,effectively integrates various information characteristics,excavates the causal relationship between events and various types of information,and realizes early prediction of hot events while ensuring time efficiency.2.For the event propagation prediction task,an event propagation prediction model based on dynamic deep network is proposed.The existing propagation prediction algorithms are difficult to effectively fuse multiple temporal features and mine key information.For this reason,this thesis designs a dynamic graph neural network to extract the key text information of the time sequence.Through the method based on the combination of two-way gating cycle unit and attention mechanism,it fully extracts the tweet forwarding,user attributes and network structure information and converts them into statistical numerical features,and then fuses the text and statistical features to realize the dynamic prediction of the event propagation process.3.Complete the hot event prediction system for social networks.In order to verify the effectiveness of the algorithm,combined with the actual needs,we designed and implemented a social network-oriented hot event prediction system,which has the functions of effective analysis and prediction of hot events and user interaction,and can help relevant departments to make decisions and early warning quickly,effectively control public opinion,avoid social panic,and maintain public security.