A Method Of Extraction Typhoon Disaster Information Using Social Media Data

Typhoon disasters pose a serious threat to the lives and property of our people every year.Due to the dispersal of the disaster situation,the traditional methods of collecting disaster information have serious lags,and it is unable to meet the needs of government departments for disaster relief work in time.Social media data represented by Twitter,Facebook and Sina Weibo can effectively compensate for the blind spots of traditional mass media coverage because of its wide participation,multi-source communication methods and timeliness of information.At the first time of the disaster,the public plays the role of dynamic sensors and timely releases relevant disaster information through the network,which has reference value for disaster emergency decision-making.However,due to the large amount of data,unstructured,high fragmentation,sparse text features and colloquialism,how to extract the superior information from the microblog data in a timely and rapid manner is of great significance to the typhoon disaster emergency management department.Aiming at this problem,this paper proposes a microblog information extraction and visualization analysis method for typhoon disasters based on typhoon disaster domain knowledge,natural language processing technology and geographic information system.Firstly,based on Latent Dirchlet Allocation,we constructs the microblog text classification system of typhoon disaster,and designs a fast classification method of microblog text based on similar topic merging.The classification accuracy of this method in training set 75.1%;Then,for the collection of microblog text,we summarizes the characteristics of each disaster based on the domain knowledge of typhoon disasters,and expands the feature vocabulary through the word vector model to achieve the specific disasters.Finally,we analyzes the expression characteristics of spatio-temporal information and typhoon disaster information in Weibo text,proposes a disaster information representation framework for typhoon disasters,and constructs a corresponding labeling system,which combines conditional random field model to identify spatio-temporal information and disasters information.It is verified that the accuracy,recall rate and F value of the information extraction system in the training set are 90.3%,67.4% and 77.2%.In addition,we use the TF-IDF algorithm to aggregate information on structured microblog data in three dimensions: space,time,and text category.Based on the above method,this paper constructs a Typhoon Disaster Information Extraction and Visualization System(TDIEVS).Taking the typhoon "Meranti" on the 14 th of 2016 as an example,we used the disaster news report and related literature to evaluate the feasibility of the method.The results show that the system can automatically extract relevant disaster information,and visually display the spatio-temporal distribution and development trend of the disaster through event maps and time series,which is conducive to improving the situational awareness of disaster emergency management departments and assisting decision support.