Research On Approximate Description Of Uncertain Concept

With the rapid development of information technology,the data produced in social production increase explosively.How to acquire knowledge efficiently and intelligently from the massive data is a research hotspot in artificial intelligence field.However,how to deal with uncertain information and acquire valuable knowledge from it is a key issue in knowledge discovery.Rough set theory,as a mathematical tool,can deal with uncertain problems effectively.In recent years,rough set theory has become an important method for dealing with uncertain information because it need not provide any prior knowledge.The upper approximation set and the lower approximation set of rough set model play pivotal roles in the description of uncertain concept.The relationships of upper approximation set and lower approximation set between two uncertain concepts in different granularity spaces will be discussed in this thesis so that the uncertain problem can be described better.In addition,combining with practical problems,a cost-sensitive approximation set model of rough sets is constructed for the approximate description of uncertain concept.The main researches of this thesis are presented as follows:(1)The relations of the upper approximation set and the lower approximation set between two uncertain target sets are described by rough equality and rough inclusion.In this thesis,some change rules of rough equality and rough inclusion between two uncertain target sets are studied in multi-granularity spaces.Then a new method is proposed to measure the similarity degree between two roughly equal target concepts,this method can not only reflect the similarity degree between two target concepts which are rough equality but also reflect the granularity of knowledge space.In addition,by using the existing knowledge granules,the optimistic ?-approximation set model and the pessimistic ?-approximation set model are proposed from optimistic view and pessimistic view respectively that used for the approximate description of two roughly equal target concepts.Moreover,some properties of the presented models are discussed in detail.(2)The cost information exists in data widely.Firstly,the necessity of considering cost information is analyzed when the approximation set of uncertain concept is constructed in this thesis.Then an approximation set model based on misclassification cost is constructed from the perspective of cost-sensitive,and some properties of this model are also discussed in detail.In order to find a suitable granularity space to describe an uncertain concept approximately that can minimize the sum of misclassification cost and test cost as far as possible in multi-granulation spaces,the contribute rate of attribute cost is defined,and then the cost-sensitive granularity optimization algorithm is proposed based on the contribute rate of attribute cost.Finally,the results of the experiment show that the proposed algorithm can be used in existing cost cognition scene,and a reasonable hierarchical granularity space and the approximation set of the uncertain concept can be obtained under the given cost scene.