Research On Theory And Method Of Knowledge Discovery Based On Formal Concept Analysis

With the development and commercialization of technologies such as big data,cloud computing,the internet of things,and artificial intelligence,data gradually presents characteristics such as huge scale,various types,low value density,and dynamic explosive growth.It has become a hot topic in the field of data mining to designing efficient dynamic knowledge discovery models for extracting key information from large-scale dynamic data sets.The formal concept analysis-based knowledge discovery aims to mine effective information from data sets based on concepts.To meet the analysis requirements for multiple data sets,researchers have proposed various formal concept analysis-based knowledge discovery models.However,existing models face the problem of low efficiency when dealing with large-scale data sets.Based on this,this paper focuses on incremental learning to carry out the theory and method of attribute reduction in the field of formal concept analysis.The details are as follows:(1)In the formal context with attribute updating,the update mechanisms of granular reduct are studied.By analyzing the extent changes of concept,the decision conditions for updating granular reduct are given.Moreover,the attribute identification degree is defined to measure the identification ability of attributes to an object based on attribute identification ability,and the attributes with higher identification degrees are preferentially selected in the reduction updating process to generate a minimal consistent attribute set.Experimental results show that the proposed attribute reduction algorithm can greatly improve the computational efficiency of reduction.(2)Based on the formal decision context,the incremental and decrement learning mechanisms of granular reduct are studied under the strategy of object updating and attribute updating.The update law of granular reduct is depicted by introducing inconsistent set and inconsistent factor,that is,if the inconsistent set is empty,the granular reduct is unchanged;otherwise,it needs to be updated.The granular reduct updating problem can be transformed into adding additional attributes to make the inconsistent objects consistent.From this point of view,the equivalent conditions for object inconsistency are given,which can be used to select attributes with pertinence and to design the incremental algorithm for granular reduct.The experimental results verify the effectiveness of the proposed algorithm in reduction calculation.(3)For the formal decision context,a dynamic rule-based classification model(DRCM)is proposed.By introducing incremental learning into the formal decision context that takes label information as decision attributes,the updating process of granular rules is depicted by using the incremental mechanism of attribute reduction.Furthermore,a rule-based classification model is designed on base of the granular rules generated.Experiment results show that DRCM can not only achieve good classification ability but also significantly improve the efficiency of rule acquisition algorithms.(4)For the fuzzy formal decision context,a fuzzy rule-based classification model(FRCM)is proposed.Firstly,the incremental mechanism of granular reduct in the regular fuzzy-crisp formal decision context with increasing objects is discussed,and an update-based granular reduct calculation method is proposed.Secondly,a novel classification model FRCM is designed based on the maximum rule similarity by cross updating granular reduct and granular rules in the incremental learning process.Experimental results show that FRCM can effectively classify fuzzy data sets and has good incremental learning ability.The above findings indicate that the proposed incremental knowledge discovery methods are of theoretical and practical important,and also provide a new approach for the research of concept continual learning.