Research On Civil Unrest Event Detection Based On Frequent Subgraph Mining

With the rapid development of the Internet,the traditional news media in this information explosion times grow fast too.The GDELT global news event database in such an increasingly developed situation is also officially released in 2013.It records since 1979 from around the world and conflict events associated with the Internet and emerging media explosive growth,released the latest global news events to the world to update every 15 minutes.With such a large number of high quality data,how to use data mining and machine learning method as the main method,put forward to reveal the correlation signal and the mechanism contained in the data and study the accuracy and generalization ability model according to the near real-time update of GDELT database,which provide a theoretical basis for the model of real-time detection and monitoring of mass protests,is an important research topic.Based on the above research objectives,the following work has been carried out:(1)The design of multi thread GDELT data acquisition system based on metadata queue,can simultaneously complete historical data collection and update the data collection of GDELT data,to ensure the integrity and real-time GDELT data,data integrity has laid the foundation for further study.(2)Design method of GDETL data warehouse based on high performance ETL.Through this method,the data query performance of GDELT data warehouse in different scenarios is increased by 3.4 times to 20 times,which provides a solid data support capability for the research of massive GDELT data.(3)An innovative approach to feature extraction of mass protest events based on frequent pattern mining is proposed.The use of labeled undirected graphs for modeling abstract data according to the time window after the GDELT event,the independent record of events into the event interaction between entities atlas,and use frequent subgraph pattern mining which can reflect the mass protests can be distinguished from the historical data mining features.(4)A model for the detection of mass protests based on frequent sub graph features.The model was constructed according to the complete frequent subgraph feature vector based event detection problem into abstract classification problem,and has good detection accuracy and generalization of the detection performance of event detection model obtained by multi classifier optimization model and lays a theoretical foundation for the subsequent detection for real-time updating of GDELT data.