Association Rule Mining And Change Mining Based On The Concept Lattice

Along with the development of information technology, there are more and more way of data production and acquisition. And it has strong ability to process huge data. However,"Data Rich, Information Poor" become true in information management.As a powerful tool, Data mining can find useful knowledge from huge data, manage and utilize it effectively.As one of the important applications of data mining, the association rule mining aimed at huge, various kinds of data, explores the meaningful association and frequent pattern in a simple and direct way, to assist data user understand the relationship between attributes clearly.In the actual database, attribute has big difference levels, and the concept lattice of the formal concept analysis can denote data on multilayer, multiple contact, which cope with the reality. So, concept lattice becomes the effective method of association rules mining.In the actual database, attribute has difference levels, and the concept lattice of the formal concept analysis can reflect data on multilayer, multiple relationship, which cope with the reality. So, Concept lattice is an efficient tool for data analysis and rule extraction from multidimensional space. The construction of concept lattice is the hot spot in the research of today.This paper studied the association extraction and change mining based on the concept lattice. It combined with three procedures:the construction of concept lattice, association rule minig and change pattern analysing.1. With the survey of main construction method of concept lattice, a fact is found that to construct the concept lattice, the distribution method has no idea on the distribution feature of context, during the division procedure. To realize this, this paper gives a context division method based on the spasity, provides two division strategy of context:horizontal and vertical, analyses the scope of application, give combination strategy accordingly, and processing method of five probably situations occurred during the combination. Finnaly, an divide-and-conquer angorithm based on sparsity is described, and its efficiency is analyzed.2. Traditionally, the minimal support factor is set by annual during the extraction of association rules from concept lattice. But using a single min_sup may get improper association rules, especially for data items which have complex distribution. To set the proper min_sup, this paper proposes an adaptive support factor and some parameter used to fixed it. Finally, three rules are described for extraction of redundant rules from concept lattice, and a NARMC algorithm is proposed.3. After mining of association rules from concept lattice, this paper solves change mining problem at last. Association rule mining give user association rules of different periods. Change mining may discover the trends of change. This paper introduced theory of change mining, proposes three parameters to judge change patterns:using similarity to judge emerging patterns, add patterns and perished patterns; using unexpectness parameter to judge unexpected conditional changes and unexpected consequent changes. An adaptive method is provided to set Rule Matching Thredshold (RMT), which can avoid wrong ajudgement on add patterns and perished patterns. Finally, an experiment on check up data shows that the discovered changes can represent the trends in the dynamic environment.