Emotional Classification Of Weibo Public Opinion Texts During The COVID-19 Epidemic

There are a large number of active users on the social networks owned by well-known domestic and foreign companies every day,constantly generating new content and accumulating a large amount of data.As the first website in The Domestic Portal to provide Weibo services of attention and communication based on interests,information and content,Sina Weibo is an important channel for public opinion dissemination.Most of the data in Weibo is mainly textual,reflecting users’ views and feelings about specific social events or things,and expressing users’ emotional tendency.As a major public health emergency affecting the world,COVID-19 has an important impact on the production and life of all mankind,as concerned with social stability and development,and also poses an unprecedented challenge to China’s social governance system.In the war of the epidemic prevention and control,Weibo is an important channel for the release of public information on the epidemic as an important part of the new media of government affairs.Analyzing the emotional tendency of public blog text is of great guiding significance for understanding the real-time emotions of the people at important moments and the aspirations of the people,in order to discover problems,analyze the causes and channel emotions,and can also provide direction for the government and departments to formulate corresponding measures.Based on the comprehensive theoretical analysis and empirical analysis,this thesis starts by introducing the mainstream development direction of applied models in the field of natural language processing and emotion classification,and sorts out and derives the commonly used theories and techniques.In the empirical section,based on 100,000 pieces of manually labeled data in the "Netizen Sentiment Identification During the COVID-19 Epidemic" competition held by the Beijing Municipal Bureau of Economy and Information Technology,the CCF Big Data Expert Committee and the Information Retrieval Professional Committee of the China Chinese information Society in early 2020,the data was cleaned using regular expressions and jieba word segmentation tools,and then the distribution characteristics,trend changes and keywords of public opinion at the time of the initial outbreak of the epidemic were explored by visual means and corresponding tools in jieba.According to the development direction of machine learning models,deep learning models and pre-trained models in natural language processing tasks,this thesis compares the training results of logistic regression,bidirectional LSTM networks,and ERNIE,which is an open-source model from Baidu on the dataset in CPU environment.The data exploration results of this thesis are instructive for analyzing the characteristics of public opinion at the time of the initial outbreak of the COVID-19 epidemic.Based on the step-by-step optimization process and final results predicted by models in this thesis,Chinese semantic pre-training models such as ERNIE perform better than traditional deep network models in the field of natural language processing,and the deep network models do better than ordinary machine learning models.