Research On Typhoon Disaster Situation Awareness Method Based On Micro-Blog Data

Typhoon is one of the major disastrous weather systems affecting China.It is of great practical significance to timely obtain disaster information,perceive the development trend of disaster situation,and quickly detect the temporal and spatial distribution pattern of losses in the process of disaster,so as to win time for the typhoon emergency management department and reasonably dispatch the limited rescue forces and materials.In the context of big data environment,the emergence of social media such as micro-blog breaks the time and space constraints of traditional information dissemination.The real-time nature of information release and the wide participation of the masses make it an important way for people to release and obtain information.However,due to the spontaneity and independence of the release source,social media data is prone to a series of problems,such as low degree of standardization,high sparsity of context information,serious colloquialism and networking,which makes it difficult to be directly applied to disaster emergency management tasks.For the purpose of typhoon disaster situation awareness research,based on the microblog data,this paper combines the advanced natural language processing technology with the disaster information characteristics in the field of typhoon disaster,integrates the methods of Theme Evolution,text classification,information extraction,information quantification and spatial analysis,and carries out research from both qualitative and quantitative aspects to mine and release the utility of micro-blog data in the field of typhoon disaster.Further build a typhoon disaster situation awareness system to realize the flow of disaster information extraction and the visualization of disaster development process,in order to help disaster emergency managers timely grasp the spatial distribution pattern of disaster development situation,reasonably formulate disaster relief strategies and allocate limited rescue forces and materials.The main research results of this paper are as follows:(1)Typhoon development situation awareness and disaster classification based on dynamic theme modelBy establishing a dynamic topic model,we can mine the topics that people are interested in during the disaster from the disordered micro-blog text,generate the topic clustering results in an unsupervised way,detect the evolution of the topic and the discussion center within the topic from the perspective of topic evolution and word evolution,and perceive the development trend of the disaster from the change of public point of view.At the same time,aiming at the lack of context semantic information of probabilistic topic model,Bert pre training model is introduced to further optimize the topic classification results.This method not only realizes the perception of typhoon development situation,but also has high classification accuracy,which lays an important foundation for subsequent research.(2)Disaster information extraction method based on rule and dictionaryIn view of the rich expression of disaster information in micro-blog text and the lack of public marked corpus,which makes the method based on machine learning difficult to migrate,this paper extracts the text for manual marking,classifies and summarizes the fine-grained lexical collocation rules of disaster information,fully considers the importance of text grammatical structure to information extraction,and constructs the constraint rules of negative words and degree adverbs;At the same time,the disaster dictionary and mapping code in the field of typhoon disaster are constructed,and the unregistered words are automatically screened,which improves the granularity of information extraction and deepens the application of information extraction method in the field of disaster.(3)Put forward the quantitative research method of disaster informationThe purpose of disaster information extraction is to conduct quantitative research,build the initial damage scale of affected entities and the weight table of disaster degree descriptors,and realize the quantification of disaster information.In addition,considering the influence of negative words and degree adverbs on the text content,the disaster value calculation method is proposed,and the disaster value of each disaster information text is calculated.Integrating the spatial analysis method,the disaster response index is proposed to re integrate the scattered micro-blog disaster information,so as to realize the overall perception of the disaster distribution pattern.(4)Construction of typhoon disaster dynamic perception and disaster analysis systemThis paper introduces the design and implementation of typhoon disaster dynamic perception and disaster analysis prototype system,expounds the overall architecture of the platform,and introduces the requirements,system architecture,development environment and the function and implementation of the subsystem in detail.The experiment shows that the disaster response index can effectively reflect the spatial distribution pattern of the actual disaster,and makes a positive exploration to further provide a scientific reference for typhoon disaster emergency management.