Research On The Construction Method Of Multi-Granularity Concept Lattices

In the background of the explosive growth of data,how to extract knowledge from huge data is crucial.Professor Wille R.from Germany proposed the formal concept analysis theory in 1982.Through the visual hierarchy of the concept lattice,the membership relationship between objects and attributes in the two-dimensional table is expressed,from which knowledge representation and rule mining are carried out.The selection of attribute granularity plays an important role in the process of concept lattice construction and knowledge discovery.And attributes of traditional concept lattice are generally fixed.Combined with the granular computing theory,the attributes can be used as the objects of granular computing.With the multi-granularity discussion of the attributes,concept lattices with different structure can be induced.Not only can the amount of concepts be controlled,but also the users can obtain knowledge based on different levels correspondingly.Therefore,it is of great value to study the transformation among attribute granularities for different kinds of concept lattices.At present,there are only the studies about the construction algorithm for multi-granularity classical concept lattice.Therefore,this thesis investigates the construction method of multi-granularity concept lattice for other different kinds of concept lattices,and proposes the corresponding Zoom algorithms: Zoom-in algorithm used to refine the attribute granularity and Zoom-out algorithm used to coarsen the attribute granularity.The main work of this thesis is as follows:1.Firstly,analyze the characteristics of object-oriented concept,attribute-oriented concept and the order relationship among concepts.Secondly,the transformation algorithm between object-oriented concept lattice and attribute-oriented concept lattice based on the same formal context is proposed.Thirdly,according to the implication relationship between different granularities of the same attribute and the relationship between the selected attribute and the intent of the concept,classify the concepts and use different calculation methods for different types of concepts in subsequent algorithms.And the construction algorithms(ORZoom and ARZoom)for multi-granularity object-oriented concept lattices and multi-granularity property-oriented concept lattices are proposed.Then,the calculation processes of ORZoom and ARZoom are demonstrated by examples.Finally,carry out experimental analysis;2.Different from the value of 0 and 1 in the traditional formal context,the formal context with real-valued has more practical significance.For the classical generalized one-sided concept lattice based on the real-valued formal context,the properties and characteristics of different concepts are discussed firstly.Secondly,the corresponding multi-granularity generalized one-sided concept lattice construction algorithms(GOSZoom)are proposed.Thirdly,examples are used to represent the impact of attribute granularity changes on the concept and concept lattice generation,and demonstrate the implementation process of the algorithms.Finally,GOSZoom algorithms and classical generalized one-sided concept lattice construction algorithm(GOSCL algorithm)are compared by experiments;3.Different from the classical generalized one-sided concept lattice,this thesis proposes multi-granularity positive generalized one-sided concept lattice construction algorithms(PGOSZoom)for positive generalized one-sided concept lattice with multiple attributes’ granularity changing simultaneously.Finally,algorithms are demonstrated and analyzed experimentally.