User Sentiment Tendency Aware Based Micro-blog Sentiment Analysis Method

With the continuous development of social media platforms,more and more people are accustomed to using Weibo to express their opinions and opinions.Using artificial intelligence technology to mine the valuable information hidden in these Weibo data can better help people to be more efficient.This article mainly explores the microblog sentiment analysis method,focusing on mining more hidden information to improve the accuracy of microblog sentiment classification.Based on the deep neural network technology,this paper analyzes the text characteristics of microblogs in general,and proposes an effective microblog sentiment analysis method based on recurrent neural networks,Attention mechanism,and graph convolutional neural networks.This article introduces the research background,current situation and significance of text sentiment analysis in detail,and then introduces the related technology of sentiment analysis in detail.In response to the current shortage of research data resources,we constructed a microblog sentiment analysis data set MEDUI containing user information and a microblog data set MUT containing user information and time series information,and disclosed the data set to work in sentiment analysis.Researchers who need to consider user factors or timing factors provide new data resources.Aiming at the lack of consideration of user's emotional orientation in sentiment analysis research,we propose a new method of microblog sentiment classification based on users 'own emotional tendencies,and describe the architecture of microblog sentiment analysis model based on users' own emotional tendencies: Acquisition,abstract representation of user's emotional tendency,structure of word representation,overall model architecture and implementation details of model training.At the same time,the experimental data,experimental environment and experimental parameters are introduced,the comparative experiment is designed,and the experimental results are analyzed and summarized.Aiming at the problems of sparse microblog data and dynamic changes of user emotions,a new method for classifying microblog emotions using different types of user context information and user emotional state information is proposed,and the model architecture is described: word representation,microblog text Characterization,userhistory information application,overall model architecture and implementation details of model training.At the same time,the experimental data,experimental environment and experimental parameters are introduced,comparative experiments are designed,and the experimental results of each module unit are analyzed and summarized.Aiming at the problem that word representation lacks global features,a novel word structure method is proposed,that is,a word representation method that includes semantic features,emotional features,and global dependencies.Based on the text graph built on the entire corpus,a graph convolution network is applied to mine the global semantic dependence of words.At the same time,the experimental data,experimental environment and experimental parameters are introduced,and the comparative experiment is designed.The experimental results are analyzed and summarized.