Research On Geographic Information Extraction Based On Social Network Data

With the rapid development of Internet technology and the popularity of intelligent mobile terminals,social networks have gradually integrated into people's daily life and become the most important communication and sharing platform among people.There are a lot of valuable information in massive social network data,including character,time,location,event and emotion,etc..The structured and unstructured data of social networks contain a large number of explicit and implicit geographic information,which plays an important role in providing geographic information services and building geographic knowledge graph.The geographic knowledge graph based on social network data can not only provide geographic information services,but also describe the behavior trajectory of specific targets,such as characters and events,which can provide data support for public opinion monitoring and analyzing,emergency monitoring,etc..The main studies and contributions of this thesis include several aspects as follows:(1)Social network data mining method based on geographic information extraction task is studied and implemented.This thesis studied the characteristics of social network data,analyzed the characteristics of geographic information extraction task,and finally proposed a data mining method for the test of geographic information extraction task.According to the method,we crawled about 20 million Tweet data from Twitter and construct a high-quality social network set.(2)A geographic information extraction model based on social network data is constructed.According to the features of social network data structure,we constructed a model for geographic information extraction task by taking advantages of device data and content data.The effect of the model on the extraction of the geographic information of the content data part is improved by using deep learning technologies.(3)The model extraction effect is analyzed experimentally.In this thesis,two groups of experiments are designed to verify the rationality of each part of the model and the effect of the deep neural network and multi-attention mechanism.The experimental results show that in the data set constructed in this thesis,each part of the model can effectively extract geographical information and the extraction accuracy of the model is more than 30%.The deep learning technology greatly improves the extraction effect of the model and the extraction effect of the model with multi-attention mechanism is better than those of model with single-attention mechanism or model without attention mechanism.