Real-time Mining And Analysis Of Urban Waterlogging Disaster Information Based On Social Media

With the increase of extreme weather and the acceleration of urbanization,urban waterlogging occurs frequently in China,which has seriously affected the production and life of urban residents and the sustainable development of urban economy and society,posing a huge threat to the safety of people’s life and property.At the same time,the traditional way of collecting disaster information is inefficient and information transmission lags behind,which cannot meet the needs of government departments to carry out emergency rescue in time.Social media has been more and more applied to disaster monitoring and obtaining disaster information due to its extensive participation,rich spatial and temporal dynamic information,large amount of network information and strong timeliness.Based on social media big data,this paper conducts theme mining and analysis on urban waterlogging disaster information to provide decision support for urban flood control and disaster relief emergency management.Firstly,this paper analyzes and designs the extraction scheme and text pretreatment process of urban waterlogging related Weibo based on keyword crawler,and establishes the urban waterlogging corpus.Then use the LDA subject mining model first topic modeling analysis of corpus,established the theme-word matrix and implied theme clustering,then use support vector machine SVM algorithm for training,build subject classification model of urban waterlogging disaster information,can be classified to extract the waterlogging disaster event information,It is helpful to further identify the related information of disaster loss,flood control and rescue,and the impact of waterlogging.At the same time,the named entity identification method and the LAC lexical analysis model developed by Baidu are adopted to identify the waterlogging address entities contained in the text from the disaster microblog data,which can make up for the shortcomings of traditional fixed-point monitoring methods and find the waterlogging areas that may have hidden dangers.The experimental results show that the accuracy of LAC model for waterlogging point identification is 82%,and the identification result is more reliable.Finally,the theme and spatiotemporal analysis of the waterlogging event was carried out by taking the Wuhan municipal waterlogging event of 7.6 in 2016 as an example.The analysis shows that the waterlogging points extracted from Weibo data in this paper roughly coincide with those published by the government.Waterlogging in Wuhan is prone to occur under overpasses,underground passages,bus stations and other low-lying areas around lakes,and there is an obvious correlation between the development trend of waterlogging events and the change trend of disaster data.Analysis result is helpful to relevant government departments to carry out disaster relief work,can be more intuitive understanding of the urban waterlogging disaster of public opinion and waterlogging events development process,have greatly enriched the shortcomings of traditional hydrologic monitoring data,the use of big data and machine learning technology for the urban waterlogging control and emergency management provides a new way.