Research Of International Relation Network Based On Text Mining

With the rapid development of communications and information technology, especially the emergence of Internet, information travels faster and easier than ever before. Internet extends the range and scope of communications in the world. Countries become closer and form a huge and complex international relation network. Based on the background, researchers hope to study international relations through the view of network, analyze the structure of international relation network, understand effect of different relations on international situation and reveal the evolution of international relations. It is called the research of international relation network which arouse more and more attention recently.One thing should be pointed out that network approach is a formalization method. It has some natural defect such as loss of information while facing with the non-formalization problem of studyinginternational relations. The network approach cannot solve all the problems ofinternational relationsas as well. Therefore, the main target of our research is not offering a package that can solve everything. We try to innovate based on current work and provide a better formalization approach to deal with international relations. Our research can offer a tool to assist researchers of international relations. On the other hand, it provides an approach to help average users getting an intuitive impression about international phenomenon. Through surveying of current work, we find that there are some defects in the constructing of international relation network which is the basis of related work. Current networks are generally constructed on structured data through manual methods or semi-manual methods. It limits the source of data, consumes too much manpower and time, and also cannot ensure timeliness and consistency. In order to solve the problem, we design method that can construct international relation network from unstructured data of texts with the help of text mining technology. The method achieves recognizing countries and their relations automatically through analyzing of texts. It expands the data source of international relation network greatly and also provides an efficient way of acquiring knowledge from information which can be used to solve the problem of information explosion. The major innovations of this thesis can be summarized as follows.(1) The method of constructing international relation network based on text mining. As there is no way can construct international relation network from text automatically in current work, we originally propose our method of construct international relation network from large scale texts through text mining which is under the unified theory of information, knowledge and intelligence. It expands data source from structured data to unstructured texts and provide an efficient way to solve the problem of information explosion. We also built a comprehensive system including data acquisition, network construction and visual interface to realize our method. Through experiment, we verify the reliability of our system.(2) Definition and extraction of international relations. Based on related research about social network and interpersonal relationship, we point that intensity, quality and status are three important features of international relations. They affect the structure and evolution of networks in different way. We design several methods to extract these features from texts and construct the international relation network. Results of experiments show satisfied performance. Finally, we attempt to propose a new frame of defining international relations which contains five new types of relations based on the three features.(3) Recognizing sentiment of relations between entities. Sentiment between countries reflects quality of relations which is an important feature of international relations. Although some work has been done on sentiment analysis, there is little work about recognizing sentiment between entities from texts especially in Chinese. A method composed of three steps is proposed. Through entities recognition and extraction, sentiment related region detection and sentiment determination; we can obtain sentiment between entities on sentence level. We compare different algorithm based on different principle (rule and machine learning) and different related region as well. The algorithm using CRF (conditional random fields) model based on syntactic dependency tree acquires best result.(4) Design and realization of visual interface. We analyze defects in current visualization approaches of international relation network and propose a new method combined with GIS (Geographic Information System). The visual interface uses Google Maps as substrate, integrates Google Maps API and Mysql database and achieves international relation network visualization on electronic maps. The interface also provides some research functions that allow users to search interesting details and see different aspects of the international network more clearly.(5) International relation network analysis and relation prediction. We analyze the constructed international relation network through visual interface and obtain some interesting discoveries about hot spots and characters of network structure. We also try to predict the development of relations through machine learning method based on data in different time periods. Experiments shows that the precision of prediction on five types of relations can reach63.1579%and with a practical significance.