Theory And Approaches For Knowledge Inference Based On Granular Computing Model

Knowledge is the important cornerstone of human cognitive ability and is also the basic research question in artificial intelligence. With the rapid develop-ment of computer technology and Internet, data and information in various fields grow explosively, and their features are dynamic, high noise, uneven distribution, high dimension. Conventional methods of knowledge inference are used in large knowledge base which is not only time-consuming, but also difficult to effectively obtain reasonable solutions of problem.Granular computing model has feature which may simulate human intelli-gence. In solving problem Granular computing model can decrease the difficul-ty and simply problem by selecting appropriate granular. Granular computing provides an effective and feasible solution for knowledge mining and knowledge inference. At present, the research of granular computing is mainly theoretical aspect, and its application still has not been widely spread. This paper mainly studies that the theory of granular computing is feasible and validate for knowl-edge inference in very large knowledge base. It is the combination of theory and application in granular computing model.In this paper, the effective method of knowledge inference is researched in large knowledge base based on granular computing model, fuzzy rough set theory and quotient space theory.The main research and innovations are as follows:(1) In rough set theory, classic attribute reduction faces discretized attribute, but, in real life, the majority of attribute is continuous. It will lose a lot of useful information, if attribute is discretized. under the circumstances, scholars extended rough set theory to fuzzy rough theory. Based on fuzzy rough set theory, we first propose fuzzy and dynamically incremental attribute reduction algorithm which is fit to retrieve in dynamic and large scale of case base.(2) To improve the retrieval efficiency in case base, this paper presents a method of granulation by reduced attribute. The specific method of granulation is given:granulation and clustering are methods which are to sum up the knowledge. k-means is a common method for clustering. Because the selection of k center point will cause disturbance of clustering accuracy, scholars present k-means++algorithm. Cases in case base are dynamical increasing, so we present dynamically incremental k-mean++algorithm to granulate case base by reduced attribute and form case-granulation. Above all, target case retrieves the most similar case-granulation rather than matches every case in case base. Finally, target case obtains the most similar case in the case-granulation which has the maximum matching degree with target case.(3) The hierarchical structure of the quotient space provide an effective way for problem solving. Professor zhang ling and academician zhang bo present fuzzy quotient space and extend classic quotient space. Above all, the method of forming quotient space by intuitionistic fuzzy tolerance relation is presented and broader hierarchical structure is obtained in this paper. We prove the same hierarchical structure induced in two ways of composition by the intuitionistic fuzzy tolerance relations and get some interesting properties.(4) Granular computing model and quotient space theory play a key role in the optimal path selection of hypergraph(complex and large network graph). In this paper, hypergraph is granulated by superedge weights of hypergraph, further-more, hierarchical quotient space chains are formed by different weights. Finally, the complexity of path search problem is reduced by hierarchical quotient space chains and the optimal route is obtained.