Research On Method Of Generating Spatial Relations Models

With the development of information technology, the amount of information increased rapidly, accompanied with the decreasing of the value of information. It seems that the huge amount of information provides more and more convenience and profit, while the fact is that it makes people more difficult to use information. Thus, plenty of big data technologies emerge as the times require, in order to professionally handle the mass data. By getting, saving, managing and analyzing the data, big data technologies help people obtain what they really need. Spatial information, as one kind of the mass data, is applied to all aspects of human life. The big data and the analysis of spatial information help people a lot in many fields, such as news push-delivery on the Internet, the accurate advertising in e-business, GPS, the analysis in traffic flow, government rescue after the disasters, selecting places for buildings in the development of a city or a company. So dealing with the spatial relations information in the mass data is quite an important work.Spatial relationships have been widely studied for more than two decades. They are widely applied in spatial query and spatial analysis, and quite a lot of spatial relations models have been proposed at home and abroad. The existing qualitative spatial relations models are often based on a sound algebra system. They are built by manual work according to the priori-knowledge of spatial data, and could not fully express the spatial relationships in the real world. The existing complex spatial relations models focus on certain spatial relations of spatial objects, so the models perform poor for the variety of spatial relations. It is necessary to analyze the priori-knowledge of spatial objects when manually building spatial relations models based on sound algebra systems, and there are many elements to be considered for spatial relations in the big data, which make it too complicated for people to formalize the mathematical system.In order to solve the problems mentioned above, we did a research on the methods of reasoning spatial relations, and a new method for automatically generating spatial relations models is proposed in this paper. We use qualitative spatial reasoning combined with machine learning, instead of building an algebra based system to set up spatial relations models. The proposed method could generate spatial relations models without any priori-knowledge, and the generality of the method could reach the requirement of the expressions for spatial relations in the big data times.The main content of this paper are as follows.(1) This paper first elaborates the background, significance and current status of researches in today’s era of spatial relations, which is a popular topic in spatial reasoning. Then, the content of the study and the structure of this paper are introduced.(2) The basic concepts of spatial relations and the relative knowledge of machine learning are introduced in this paper, followed by descriptions of a machine learning algorithm named MLNB.(3) By analyzing the problems in the existing spatial relations models, we put forward a method for generating spatial relations models, in which a general feature set for spatial relations is proposed. The procedure of this method is mentioned in this paper, together with the detailed explanations for the division of spatial objects, including the division of sub-regions and extended-regions. We also proposed a multi-levels strategy for feature selection on this basis, to reduce the number of features to be conducted in the following process by grouping the sub-regions into several levels.(4) The method proposed in this paper is then used in experiments in several datasets. Furthermore, the method is applied to an application of recognizing spatial relations in Chinese texts. By analyzing the results of the experiments, we can conclude that the proposed method could perform well in different circumstances.(5) At the end of this paper, we presented the summary of this paper and looked forward to the further work.