Research And Implementation Of Monitoring And Analysis Method For Network Media Event

Due to the rapid development of network information technology,network media has penetrated into every aspect of public life.Compared with traditional media,the influence of network media tends to spread more rapidly,and the amount of information is larger.Meanwhile,the authenticity of network media cannot be strictly controlled,and the content is mostly presented in the form of short texts.In the face of massive online media data,if relevant departments fail to respond to rapidly spreading social events in a timely manner,adverse or even false information will get a lot of attention and spread,accelerate the deterioration of social events,cause social panic,and affect the harmonious development of the country.To sum up,how to strictly control the authenticity of network information,quickly extract the topic categories of social events,and accurately analyze the public's preference for a certain event is of great significance to the monitoring and analysis of public opinion.This thesis studies network media event data from three aspects,the specific research content is as follows:1)This thesis proposes an event authenticity detection method combining convolutional neural network and bidirectional gated loop unit for monitoring media event content.Firstly,the convolutional neural network is used to learn the grammatical features of the event text after quantization of the media event report.Secondly,the obtained features are spliced and input into the gated loop unit to learn the sequential feature representation of the event,and finally achieve the purpose of event authenticity detection.This method combines the advantages of the two network models and achieves the purpose of simultaneously learning the event text syntax and event sequence related features of network media.2)This thesis proposes a short-text topic discovery method combining BTM and weighted K-means to quickly analyze topic categories of social events.Firstly,the BTM model is used to model the short text in media data to obtain the subject words of text content.Secondly,the weighted K-means clustering algorithm is used to assign different feature weights to each attribute feature to realize the short text topic discovery of media events.This method not only solves the problem of media data sparsity,but also avoids the defects of traditional K-means algorithm.3)This thesis proposes a public tendency analysis method based on self-attention mechanism and bidirectional long and short-term memory neural network to accurately analyze the trend of public opinion.Firstly,a model is built to learn the semantic information in the text data.Secondly,the weight of each word is assigned according to its emotional tendency.After obtaining the weight distribution of words,the weight of key words in the data is enhanced to realize the public orientation analysis.This method weakens the influence of irrelevant words in data text on propensity analysis and improves the analysis effect of the model.Experimental results show that the three models have good performance.For network media data,the F1 value of event authenticity detection based on CNN extracting grammatical features and BI-GRU algorithm extracting sequence-related features reaches about 97.12%.Combining word co-occurrence topic model and weighted K-means clustering,the topic consistency score of the analysis algorithm reached-48.54.Using the analysis model of attention mechanism and BI-LSTM network,the weight distribution of each word to emotional tendency was considered,and its F1 value could reach about86.60%.In the end,the problems existing in the research work and the next research work are explained.