Research On Multi-level Concept Construction Base On Granular Computing Method

Ontology learning is a research focus in the artificial intelligence field. At present, it has been a great development. Due to the concept learning proposed, the chaos of concepts in the areas system is reduced; the concept of sharing and interoperation in many different fields are realized by concept study; the problem of concept sharing and concept reuse are resolved. As a basic of Ontology learning, Concept learning is widely used in knowledge engineering, digital libraries, information retrieval and many other fields.Granular computing is an emerging conceptual and computing paradigm in information processing. Just as a great umbrella, it may be regarded as a label of theories, methodologies, techniques, and tools that make use of granules. As computing units, granules can decompose a complex problem into some simple or small problems, so that the computing costs are reduced, a problem can be understood better, and the trivial can be avoided during problem solving. It has been applied in many areas like datamining and machine learning.Based on analysis of basically granular computing theory and concept acquisition theory, granular computing is applied to multi-level concept construction, and the arithmetic of multi-level concept construction (CGS) is put forward, which is based on granular computing. The main ideas of the method are: Abstract concept is described by information granules, and Based on granular computing multi-level Granularity Structure is generated so as realize the multi-level concept with different abstract level. This dissertation foucuses on study Multi-level Concept Construction from static and dynamic aspect, and the main contribution is as follows: First, using granular computing theory, the formal of information granules in the Domain-specific Concept, concept granules definition and the basic arithmetics of concept granules are put forward, and the construction arithmetic of multilevel concept granular space is designed; Second, Multi-level concept granular space is constructed in the field of ZOO data by the CGS arithmetic, so as obtain the different layers concept; Third, the efficiency of the CGS arithmetic has been testified by comparison experiment between CGS arithmetic and kmeans arithmetic. Fourth, According to the change types of dynamic data, the arithmetic of multilevel concept incremental learning is put forward, and Multilevel concept incremental learning algorithm demonstration System based on JAVA is put forward. Finally, comparing the incremental learning algorithm with no increment learning algorithm, the result show that incremental learning algorithm is of good performances in the time complexity and spatial complexity.Based on above fruits, we have a conclusion about the theory and applications of concept granular spaces. Besides, some problems in the model and our future works are proposed.