Research On The Sentiment Analysis Of Microblog Texts During The COVID-19 Epidemic

With the rapid popularization of the Internet and the vigorous development of related technologies,microblog has become a major tool for more and more people to obtain daily information and share personal opinions with its outstanding advantages of freedom and flexibility.Most of the content produced by users on microblog platforms is presented in the form of text,and these microblog texts can often directly reflect the current sentiment tendency of their publishers towards a specific object(such as an event,a product,a service,etc.),and then you can understand their views,attitudes,positions,etc.towards the object.With the outbreak of COVID-19 epidemic in China,many microblog texts about the epidemic have appeared on the domestic microblog platform,and many scholars have conducted a lot of research on them.As a hot research direction in the field of Natural Language Processing(NLP),Sentiment Analysis based on microblog texts has therefore become a research topic that scholars pay attention to in this special period.In the research of this paper,Word2vec-HAN,a model for sentiment analysis,is constructed by using Word2 vec word vector,Gated Recurrent Unit(GRU),Attention Mechanism and other deep learning techniques on two open source microblog data sets of COVID-19 epidemic,and the effectiveness of Word2vec-HAN is proved by multi-model comparative experiments on the test sets divided by the two data sets respectively.Specifically,the experiment first trains Word2 vec word vector on a large Chinese microblog corpus and establishes the corresponding embedding layer,then connects the layer with a double-layer network,which is composed of GRU and Attention Mechanism,and finally obtains the sentiment tendency of microblog texts through the Softmax Function of an output layer.Experimental results show that compared with other models,the prediction performance of Word2vec-HAN is the best on both data sets.Additionally,in order to intuitively illustrate the interpretability of Attention Mechanism in Word2vec-HAN,some visualization methods are used in the experiment to present the distribution of attention weights at the two levels of word and sentence.