Research On Concept Learning Based On FCA

With the development of information technology,there are many ways to get data,such as television,newspapers,Internet,etc.The cycle of data acquisition is also decreasing.Faced with massive structured,unstructured,semi-structured data,how to quickly and effectively mining the potential value from these data is the current research hotspot,at the same time,it is also a challenge and opportunity for machine learning.Rough set theory was proposed by Pawlak in 1982,and it is widely used in data mining,machine learning,decision analysis and other fields.Formal concept analysis(FCA)is a mathematical tool in knowledge discovery proposed by R.Wille in 1982.It is widely used in data mining,clustering,classification and so on.Rough set is mainly used in the knowledge representation of uncertainty.Formal concept analysis is mainly to explore the intrinsic relationship between a class and the properties of the class.The combination of rough sets and formal concept analysis can exploit the uncertain relationship between a class and attributes of the class.In the era of multi-channel access to information,information obtained from a single information source is often fuzzy or incomplete,so it is necessary to fuse informations which they are from multiple information sources.The purpose of fusion is to integrate the information from multiple information sources into a larger information body.So that the fusion with the effect of "1 +1> 2".This thesis is the main feature concept learning and the multi-source fuzzy concept learning based on formal concept analysis.The mathematical properties of the main features and the information fusion are studied,and the fusion algorithm and the fuzzy concept learning algorithm are designed.The effectiveness of the proposed method is verified by numerical experiments.The main innovations are as follows:1.In the formal context,the ? main features are defined,and the difference between the inevitable features and the likelihood features is expounded in the mathematical language.The influence of the basic ratio on the cognition in the formal context is studied.The relationship between the basic ratio and the ? main features is defined to be the credibility.How to express the degree of credibility of the object with a certain attribute(feature)belonging to the concept is studied,and the relationship between the credibility and the main features is discussed.Finally,through the calculation of the case,the credibility is proved to be better than only the single main feature in the problem of whether the object belongs to the concept,and the validity and feasibility of the model are verified.2.Conditional entropy is a kind of measurement of information.The information fusion based on conditional entropy can improve the quality of classification while reducing the redundant information,so that the similarity class is finer.The difference of the objects of the same category is small;the difference of objects of the different category is larger.The fuzzy concept is obtained on the basis of the fusion.And according to the proposed the approach of conditional entropy fusion design the corresponding conditional entropy fusion algorithm.The two kinds of fuzzy concept learning algorithm based on object information and attribute information is designed,a series of numerical experiments is designed by use the UCI data set.The conditional entropy fusion proposed by this thesis is compared with the traditional mean fusion;it can improve the quality of classification with reducing the redundant information.The mean fusion is only the use of statistical methods to compress the multi-source information and can not improve the quality of classification.