Research On Network Information Extraction And Visualization Technology Of Incident

Whenever an unexpected event occurs,the relevant information will quickly occupy the top of the search list on the Internet.Obtaining relevant key information in a timely manner from a large amount of network information has become an urgent problem,and has led to the application of information extraction technology.Information extraction faces the following problems in identifying relevant information from network texts.(1)New unregistered words affect the performance of named entity recognition.(2)The attribute information characteristics of an event are difficult to obtain.(3)Event information display methods are relatively simple,and the degree of integration is not high.Deep learning and semantic dependence parsing are expected to improve the performance of information extraction,and GIS visualization can display different data together.In this thesis,the technology of information extraction and visualization for incident network information have been studied.The main research results are listed as following:(1)Named entity recognition is studied.Incident information contains a large number of newly added unregistered words,which affect the performance of named entity recognition.A Chinese named entity recognition method based on BiLSTM and CRF is proposed by using N-gram attribute of FastText word vector.Firstly,the input data is segmented into characters,and word vectors are generated by FastText tool.Then,word vectors are input into BiLSTM neural network to extract global features.Finally,according to the output feature sequence,CRF is used to select the most probable annotation sequence to realize named entity recognition.Experimental results show that this method can effectively improve the effect of unregistered words on named entity recognition and enhance the performance of named entity recognition.(2)Extraction of incident attribute information is researched.Current researches are mainly based on corpora,and existing techniques for open domain corpora are not applicable.An attribute information extraction method based on semantic dependency parsing and rule template is presented.Firstly,attribute information expression is extracted from each document to form an attribute information set.Secondly,rule templates corresponding to the expression of attribute information are generated respectively.Thirdly,semantic dependency parsing to supplement rule templates and extraction of event attribute information is achieved.Finally,the extracted attribute information is compared with the correct attribute information,and the corresponding rule templates are adjusted to improve extraction.Experimental results show that the method effectively improves extraction performance of incident attribute information.(3)GIS visualization of emergencies is explored.Chart visualization does not provide an exhaustive representation of incident information,which affects the control of the entire emergency.A GIS-based emergency visualization method is designed.Firstly,the Google Map API is used to map geographic locations to geographic coordinates.Then,attribute information is layered on a map.Finally,all hierarchical data is integrated to achieve GIS visualization.Experimental results show that the method can generate a clear visual display and effectively present the key information of emergency.