Global Social Media Monitoring Based On Artificial Intelligence

The rapid development of network social media has given rise to network public opinion.Network public opinion reflects social situations,and strengthening the regulation of network public opinion is essential to social governance.Since the 18th National Congress,the Party Central Committee has attached great importance to network governance,requiring scientific and technological innovation to help govern cyberspace,improve monitoring and early warning capabilities against irregularities,and resolutely maintain network order.However,due to the large scale of social media data,the efficiency of manual analysis and review of social media data in the past was very low.It is necessary to study the use of computer technology,machine learning,and deep learning to achieve automatic collection,early warning,and analysis of network public opinion data,to make timely responses to real-time network public opinion,and resolve public opinion crises.Due to the weak data analysis capability of massive public opinion in the traditional social media monitoring and early warning system,this paper carried out theoretical and application research around several critical technologies in network opinion data mining based on the actual scenario requirements,improving the deep automatic analysis capability of social media monitoring through various techniques of artificial intelligence,and carried out complete research and development of a social media monitoring platform based on the relevant research,which can mine more potential public opinion value of social media data.Based on the above urgent need for monitoring,early warning,and analysis of social media data,the specific research content of this paper includes the following:(1)This paper proposed a low-resource sentiment analysis method based on knowledge distillation and designed a lightweight BERT model with fewer layers and parameters than the benchmark model.The student model learns the capabilities of the teacher model by the loss calculation at the attention matrix level,the hidden state level,and the word embedding representation level.We designed various downstream task classification models based on deep learning techniques for comparison.The experiments finally achieved fast sentiment classification with low resources and the classification accuracy could reach 95%of the benchmark model.The method can predict a large amount of text data quickly and accurately with low resource consumption,and can meet the requirements of practical application scenarios of sentiment analysis for social media monitoring;(2)This paper designed a topic detection method based on hybrid text feature extraction and enhances the text features with importance indices on top of the traditional method of extracting text features using the word embedding representation training method.The application of the method to social media monitoring can effectively solve the problem that users have difficulty quickly understanding the hot public opinion situation in a short time due to the information explosion;(3)Derived from a practical task,this paper conducted thorough research on social media monitoring products in the domestic market and the requirements of three types of potential users from the perspective of commercial software products,it designed a multi-layer architecture for the construction of social media monitoring platform.This paper proposed solutions to the pain points of existing social media monitoring products in the domestic market.Based on(1)and(2),an artificial intelligence-based social media monitoring platform is built using various cutting-edge engineering frameworks and put into operation,which has realized the real-time topic monitoring and early warning functions of mainstream social media platforms such as Zhihu,Sina Microblog,and TikTok,focusing on improving the ability of the social media monitoring platform to mine multiple types of data from the massive public opinion data.It has enhanced the ability of the social media monitoring platform to mine multiple types of data from the massive public opinion data.