Data Mining and Spatial Analysis of Social Media Text Based on the BERT-CNN Model to Achieve Situational Awareness: a Case Study of COVID-19

In response to the COVID-19,social media big data has played an important role in epidemic warning,tracking the source of infection,and public opinion monitoring,providing strong technical support for China’s epidemic prevention and control work.The paper used Sina Weibo posts related to COVID-19 hashtags as the data source,and built a BERT-CNN deep learning model to perform fine-grained and high-precision topic classificationon massive social media posts.Taking Shenzhen as a region of interest,we mined the“epidemic data bulletin”and“daily life impact”posts during the epidemic for spatial analysis.The results show that the confirmed communities and designated hospitals in Shenzhen as a whole present the characteristics of“sparse east and dense west”,and there is a strong positive spatial correlation between the number of confirmed cases and social media response.Specifically,Nanshan District,Futian District and Luohu District have more confirmed cases due to large population movements and dense transportation networks,and social media has responded more violently,and people’s lives have been greatly affected.However,Yantian District,Pingshan District and Dapeng New District showed opposite characteristics.The case study results further show that using deep learning methods to mine text information in social media is scientifically feasible for improving situational awareness and decision support during the COVID-19.

Analysis of the trend of global power sources based on comment emotion mining

In recent years,renewable energy technologies have been developed vigorously,and related supporting policies have been issued.The developmental trend of different energy sources directly affects the future developmental pattern of the energy and power industry.Energy trend research can be quantified through data statistics and model calculations;however,parameter settings and optimization are difficult,and the analysis results sometimes do not reflect objective reality.This paper proposes an energy and power information analysis method based on emotion mining.This method collects energy commentary news and literature reports from many authoritative media around the world and builds a convolutional neural network model and a text analysis model for topic classification and positive/negative emotion evaluation,which helps obtain text evaluation matrixes for all collected texts.Finally,a long-short-term memory model algorithm is employed to predict the future development prospects and market trends for various types of energy based on the analyzed emotions in different time spans.Experimental results indicate that energy trend analysis based on this method is consistent with the real scenario,has good applicability,and can provide a useful reference for the development of energy and power resources and of other industry areas as well.

Stochastic Variational Inference-Based Parallel and Online Supervised Topic Model for Large-Scale Text Processing

Topic modeling is a mainstream and effective technology to deal with text data, with wide applications in text analysis, natural language, personalized recommendation, computer vision, etc. Among all the known topic models, supervised Latent Dirichlet Allocation （sLDA） is acknowledged as a popular and competitive supervised topic model. How- ever, the gradual increase of the scale of datasets makes sLDA more and more inefficient and time-consuming, and limits its applications in a very narrow range. To solve it, a parallel online sLDA, named PO-sLDA （Parallel and Online sLDA）, is proposed in this study. It uses the stochastic variational inference as the learning method to make the training procedure more rapid and efficient, and a parallel computing mechanism implemented via the MapReduce framework is proposed to promote the capacity of cloud computing and big data processing. The online training capacity supported by PO-sLDA expands the application scope of this approach, making it instrumental for real-life applications with high real-time demand. The validation using two datasets with different sizes shows that the proposed approach has the comparative accuracy as the sLDA and can efficiently accelerate the training procedure. Moreover, its good convergence and online training capacity make it lucrative for the large-scale text data analyzing and processing.