| In the context of the 5G era,big data derived from social media has become a new data source in urban disaster emergency management with rapid response,spontaneous feedback,and low acquisition cost.The role played in disaster analysis and promoting urban safety is given a new meaning.Especially in response to the COVID-19 pandemic,social media big data played a special role in public opinion monitoring,epidemic early warning and infection source tracking.However,social media big data has outstanding characteristics of massive and multi-modality,and effective information is difficult to be mined.Therefore,timely,accurate and efficient text classification has become a key technology that urgently needs to be broken through in the application of social media big data.In recent years,with the rapid development of deep learning technology in natural language processing,neural network models have been widely used in text classification tasks.However,due to the lack of a comprehensive epidemic topic corpus that can reflect the content characteristics of domestic social media text and the current situation of epidemic prevention,and the insufficient semantic feature extraction capabilities of existing topic classification models,it is difficult for the current deep learning-based topic classification methods to dig out the fine-grained and highprecision epidemic topic information in line with the current situation of epidemic prevention in China from the massive multimodal social media posts,which makes it difficult to provide effective data support for the follow-up epidemic emergency management.In response to the above problems,an improved ERNIE pre-trained language model was constructed in this thesis to achieve more efficient mining of fine-grained and high-precision epidemic topic information from social media big data,and based on this,domestic public opinion analysis and city-level epidemic emergency prevention and control application research were conducted.The main works were as follows:(1)A topic classification method based on the ERNIE-CNN model was designed.Ten types of epidemic thematic corpora reflecting different aspects of society was created to address the content characteristics of Weibo text content during the COVID-19 pandemic and the current situation of domestic epidemic prevention and control in China.An ERNIE-CNN topic classification model was constructed.By embedding the ERNIE pre-trained model into the CNN network,the local epidemic semantic features were integrated on the basis of the original large-scale general corpus,and the CNN network was used for optimization training and classification,so as to realize finegrained,high-precision epidemic topic information mining from massive social media text streams.(2)Analysis of the spatio-temporal characteristics of domestic public opinion based on epidemic thematic classification information.The temporal trends and spatial distribution patterns of Weibo posts were analyzed to reveal the Spatio-temporal characteristics of Weibo posts during the epidemic(2019/12/1-2020/4/30),in order to make a macro perception of the development of domestic public opinion.(3)Research on the application of city-level epidemic emergency prevention and control based on the classified information of the epidemic.Taking Shenzhen as an example,the spatial pattern of confirmed communities and designated hospitals/fever clinics in Shenzhen during the epidemic was explored in this thesis,and the correlation between the distribution of confirmed cases and the impact of residents' daily lives was further explored.In addition,the back-end database was built using the topic classification text,and a We Chat applet "Epidemo" was developed to support functions such as "epidemic around" and "help & consultation".It is shown that,(1)The ERNIE-CNN topic classification model constructed in the thesis can identify fine-grained,high-precision epidemic topic information reflecting the current situation of domestic epidemic prevention and control from the massive social text streams,and outperforms other topic classification benchmark models.(2)The results of the analysis of the spatio-temporal characteristics of public opinion are in line with the reality of the epidemic situation in the development stage of China.It is manifested as: during the epidemic,Weibo posts showed the characteristics of a " dramatic increase,then a gradual decline,and finally a slight increase" in time.Spatially,the Weibo posts are mainly concentrated in the central and eastern regions of China,and four hotspots were formed.(3)During the epidemic,the confirmed cases were mainly concentrated in urban areas with convenient transportation and densely populated areas,where residents' daily lives were more affected and the corresponding social media response was more intense.In addition,the We Chat applet could provide convenient services for city residents during the epidemic,and provide network politics and decision support services for city managers.The Weibo posts topic classification method based on the ERNIE-CNN model could provide reliable data support for the emergency management of the epidemic,and it would be scientifically feasible for improving the situational awareness during the epidemic. |
PDF Full Text Request