Visual Mining Of Multi-Valued Attribute Association Rules Based On Concept Lattice

Visualization of multi-valued attribute association rules mining is able to present thepotential frequent patterns and relationships among the multi-valued data (attributes orvariables) in database by means of visualization technology. Visualization technology can beseamlessly integrated into the process of association rules mining, which can intuitive, clearlyreveal mining results. It is more convenient and faster for us to discover the valuableinformation hidden in the data and improve decision-making efficiency. As a visualrepresentation formalization of knowledge, concept lattice has been applied to the field ofdata mining. In this paper, a complete solution of visualization for multi-valued attributeassociation rules mining with the concept lattice theory is proposed, which makes use of thenew mechanism of data source visualization, interactive parameter adjustment, multi-valuedattribute association rules mining algorithm, visualization of frequent itemset and associationrules as well as knowledge representation of association rules, so that users can replace thedomain-specific experts and carry on directly in data mining. The scheme could promote theefficiency of mining and significantly improve the availability of mining results.This paper is to study and accomplish the visualization of multi-valued attributeassociation rules mining, and to complete the following works:1. Introduced the classification of multi-valued attribute data for visualization ofassociation rules mining algorithm, and established a complete mining parameters adjustmentmechanism.According to the characteristics of a province’s whole demographical data, this paperpresented the redefinition and classification of multi-valued attribute data for the visualizationof multi-valued attribute association rules mining, which included numerical multi-valuedattribute, interval multi-valued attribute and categorical multi-valued attribute; established acomplete mining parameters adjustment mechanism based on support, confidence, keyattribute factor and concept hierarchy factor, we can mine the frequent itemset and rules byadjusting relative parameters in the whole mining process, is convenient for users to selectkey attribute values to mine and analyze rules as well as improve speed and mining algorithm efficiency.2. Proposed an improvement of Apriori algorithm based on key attribute factor andconcept hierarchy factor.Considering the troubles aroused by the traditional association rules mining algorithmswhich are lack of efficient data selection mechanism for users, especially the disadvantage ofdealing with multi-valued attribute data, this paper proposed an improvement of Apriorialgorithm based on the KAF factor and the CHF factor to mine multi-valued attributeassociation rules. From the execution speed and the mining efficiency compared with Apriorialgorithm, the improved mining algorithm has better performance.3. Presented a visualizing methodology of multi-valued attribute association rules basedon concept lattice.Due to the general association rules visualization approaches which are unable to expressthe frequent patterns and relationships of items, particularly lack of multi-schemarepresentation for association rules, this paper introduces a novel visualizing algorithm formulti-valued association rules mining. By using concept lattice, multi-valued attribute dataitems can be organically organized, so that the relationships among the data been able tointuitive reflect through the generalization and specialization relationships of concept latticenodes. Not only could this methodology help users of the frequent itemset representation anddynamic analysis, but also had been able to achieve frequent itemset visualization andmulti-schema visualization of association rules, which includes the type of one to one, one tomany, many to one, many to many and concept hierarchy.4. Introduced a novel approach for the knowledge representation of association rulesbased on conceptual graphs.Given the problems aroused by the ordinary association rules presentation formalizingapproaches which are powerless to demonstrate the domain knowledge, more than ever notconducive to express the relationships of items and the implicit information of rules, thispaper introduces a novel methodology for the knowledge representation of association rulesbased on conceptual graphs, which consists of schema definition and schema parse. These twoschemas can effectively parse the association rules into the conceptual graphs representationformalism by using conceptual graphs, so that the relationships among data items could be accurately and succinctly expressed with the help of the concept nodes and relationship nodes.The knowledge representation of association rules is closed to natural language, which hasmore strong readability for users to in-depth analysis and research association rules.