Research On Prediction Of State Conflict Behavior Based On Machine Learning

The prediction of national conflict behavior in the field of international relations research refers to the prediction and analysis of the outbreak,persistence and casualties of war,armed confrontation,political turmoil,etc,and can provide certain auxiliary support for making strategic decisions in advance.With the development of technologies such as big data and natural language processing,it is possible to predict conflict behavior based on open mass news data.The "event data analysis method" widely used in the quantitative prediction research of national conflict behavior has some problems,such as strong dependence on expert knowledge and poor timeliness.This paper comprehensively uses the relevant technology in the field of artificial intelligence to capture massive news data from the Internet for text feature extraction and construction of predictive model.Dynamic Bayesian network model is used to build the conflict prediction model between China and the other seven participating countries in the South China Sea issue.The main research contents are listed as follows:First,an automatic feature extraction algorithm based on Hierarchical Clustering,Term Contribution(TC)and Latent Dirichlet Allocation(LDA)topic models is proposed.Multinomial logistic regression is used to construct conflict prediction.model.The proposed algorithm uses hierarchical clustering to solve the problem that the LDA topic model need to determine the number of the topics.The fusion of TC and LDA can extract more representative keyword features from the massive news data.The topic vector is constructed based on the extracted topic words,and is sequentially fed into the Multinomial logistic regression model for iterative training to obtain national conflict behavior trend prediction model.The feasibility and effectiveness of the proposed algorithm are verified by the prediction of North Korean nuclear behavior and Syrian-Israeli conflict.Second,In the international events in which many countries participate,conflicts between countries are not only affected by bi-lateral relations,but also subject to bi-lateral relations between other countries.In order to construct a conflict behavior prediction model between China and the other seven participating countries in the South China Sea issue,a national conflict behavior prediction method based on Event Extraction(EE),Timing Contributions(TCs)and Dynamic Bayesian Networks(DBN)is proposed.The news data is divided by month,the feature words are extracted,and the monthly TCs is calculated according to the frequency of events.The national conflict behavior data set is constructed based on the event score table built by experts and event extraction results obtained by template matching method,and then take the constructed data set as input to train the structure and parameters of the DBN fusing TCs.The prediction accuracy of the conflict behavior prediction model between China and other participating countries in South China Sea dispute reached 75.7%.The mutual influence of conflit behaviors between different countries are analyzed.Finally,it summarizes the research achievements of this paper,and make the prospects.