Design And Implementation Of Microblog Topic Sentiment Analysis System

With the rapid development of the Internet,social media platforms represented by microblogs have been rapidly integrated into our lives.Microblog topics become the focus of users’ interests and the wind vane of public opinion.Sentiment analysis of microblog topics helps the government manage public opinion and pacify people’s emotions,enables businesses analyze market demand and improve product quality,and helps online celebrities handle crisis public relations and maintain their reputation.It’s challenging to analyse the sentiment of microblog topics due to free opinion expression,fragmented information dissemination,and the fact that some users may lead the sentiment of topics for commercial purposes.Sentiment analysis of microblog texts is crucial in microblog topic sentiment analysis.Most studies analyze sentiment on a coarse-grained level,while some scholars conduct finegrained sentiment analysis to classify emotions into multiple categories with room for algorithm improvement.In addition,most scholars use methods such as sentiment dictionaries to analyze the sentiment of topics based on emotional computing ignoring the dissemination characteristics of of microblog messages.Few scholars analyze the sentiment of topics based on social networks using explicit social relationships ignoring implicit social relationships.To address the above shortcomings,this thesis designs and implements a system which can analyze the sentiment of a certain microblog topic and discover the topic sentiment evolution law.The specific research contents are as follows:(1)A neural network model combining adversarial training and BERT＿Bi GRU is proposed for fine-grained sentiment analysis of microblog texts to classify emotions into multiple categories.In this thesis,BERT is combined with Bi GRU to extract deep semantic features,,and adversarial training is added to enhance the robustness of the model.This model shows better performance compared with other models.(2)From the perspective of emotional computation,a quantitative representation and evolutionary analysis of sentiment in microblog topics considering microblog information dissemination characteristics in the network is proposed.The method quantifies topic sentiment and tracks its change over time by combining sentiment tendency with propagation characteristics,which can discover the sentiment change,track the trend of sentiment diffusion and find the time point of sentiment outbreak.Examples of analysis confirm its effectiveness.(3)From the perspective of social network,sentiment characteristics of opinion leaders under microblog topics are analyzed based on the implicit relationship network.The method mines implicit relationships among opinion leaders,constructs an implicit relationship network of them,uses the Louvain algorithm to divide the network into communities,and embeds the opinion leaders’ sentiment tendency toward topics into the network,and judges whether they lead topic sentiment through network feature analysis.The result shows that the method can effectively distinguish whether the sentiment of microblog topics is influenced by and led by opinion leaders.(4)The microblog topic sentiment analysis system was developed.Based on the above research results,this thesis designs and implements a microblog topic sentiment analysis system,including modules for topic data collection and processing,topic sentiment analysis,and basic topic overview,which provides users with an overview of a certain microblog topic,help users analyze the trend of topic sentiment evolution,analyze the pattern of topic sentiment change.This thesis reveals hidden sentiment information and tracks sentiment evolution under different topics in massive microblog data.These findings can be applied in the fields of public opinion monitoring,market discovery,product promotion and crisis public relations,with important practical value and broad development prospect.