Research On Public Opinion Analysis And Monitoring Of Online Social Networks Based On Temporal And Spatial Characteristics

The monitoring and analysis of social topics in social networks has become a hot research direction.How to efficiently mine public opinion topics and opinion leaders from social networks is of great significance.It is urgent to study the mining of public opinion topics and opinion leaders in social networks and correctly guide public opinion guidance.As an important channel for people to spread information and express their views,social network contains a large amount of rich topic information,and has gradually become the core position of the generation,fermentation,evolution and dissemination of public opinion.Therefore,how to analyze and monitor effective public opinion information from a large number of social network topics is very important for relevant government departments to carry out public opinion monitoring and emergency response.However,the dynamic changes of public opinion information in social networks face the problem of semantic sparseness of short texts,which brings severe challenges to the analysis and monitoring of public opinion in social networks.Therefore,the research on public opinion analysis and monitoring of online social networks based on temporal and spatial characteristics has important theoretical significance and application value.Aiming at the problems existing in the existing public opinion topic monitoring methods,this dissertation analyzes and monitors the topic.From the perspective of semantic analysis,this dissertation makes an in-depth analysis of dialogue inscriptions.An emotion analysis method based on probability model is proposed to analyze the emotion of topics,which are divided into positive topics,negative topics and neutral topics.Poisson distribution method is used to simulate the change of topic over time.In the spatial location network,the topic is monitored by pruning according to the change of spatial distance,and a model simulating important users in social network is constructed.The main work of this dissertation is as follows:(1)In terms of semantic analysis of public opinion information in social networks,a semantic network suitable for short text analysis is constructed based on lexical semantic knowledge,which overcomes the defect that traditional methods can not be applied to short text.A semantic analysis framework combining semantic information and word grammar information is proposed to analyze the semantics of social network short texts.The chain model and maximum spanning tree method are used to monitor the types of words,and the weighted voting algorithm is used to eliminate semantic ambiguity.Finally,the interpretation of the text in line with semantic coherence is obtained,and the Pearson correlation coefficient(PCC)is used to describe the semantic similarity between words.The accuracy of word types is compared with chain model method and Stanford tagger method from lexical level,semantic level,word item level and query level.The experimental results show that the proposed method is better than the two comparison methods.In terms of calculating semantic similarity,by mapping words to the concept space and orthogonalizing the concept space by a clustering method,the accuracy of semantic similarity calculation between words is improved.Experiments show that the proposed method can accurately calculate the semantic similarity between words and eliminate the fuzziness and ambiguity of words.(2)In terms of topic mining and monitoring in social networks,in order to solve the disadvantage that the traditional topic monitoring methods can not accurately monitor the topic corresponding to the response time point over time,a method of dynamic Poisson time topic monitoring model is proposed.This method can find the topic conforming to the specific time point over time,and the concept of word change rate is proposed,In the variational reasoning of the model,the gradient of Poisson log likelihood and Hessian matrix are used to deduce the distribution of topics,and the variational EM framework is used to estimate the parameters in the model.In order to measure the effectiveness of the model,experiments are carried out from the dimensions of topic coherence and confusion,and the experimental results verify the effectiveness of the proposed method.(3)In terms of sentiment analysis in social networks,an emotion analysis method based on probability model(EAMOP)is proposed,which is a four-tier probability model,in which emotion tags are associated with text,topic is associated with emotion tags,and topic words are associated with emotion tags and topics,so as to realize emotion monitoring of topics from the text level.EAMOP method can simultaneously monitor public opinion topics and classify topics.The experimental results show that EAMOP has a certain improvement in accuracy compared with other complex neural network algorithms,and the accuracy of the proposed eamop model is better than text CNN model,LSTM model,multinational NB model and other complex neural network algorithms.(4)In terms of topic monitoring and opinion leader monitoring in social networks,a space-based contact network method for monitoring opinion leaders and related topics is proposed.The opinion leaders obtained by this method have high spatial relevance and high social relevance.By using the social space index,the search scope of opinion leaders is reduced on a large scale,the time complexity and spatial complexity are reduced,and the pruning method is used to reduce the search space.This method considers the semantics between opinion leaders,the interaction among users,structural cohesion,mutual influence and the number of interactions among users,reduces the search space and improves the effectiveness of monitoring.