Updating Method And Implement Association Rules Based On Probabilistic Graphical Models

The Association Rules updating is another profound impact technology on the data mining following the association rules mining. With the increasing number of database transaction, as well as the support and confidence of association rules require changing, the maintenance of the association rules presented new challenges to the researchers. In practical applications, mining association rules may exist a large number of redundant false, in this case, since the changes in transaction database or demand, to be updated all the association rules is clearly not reasonable, and often having a larger overhead.For the post-maintenance of association rules issues, especially after the transaction database changed, how to update the association rules effectively is the main problem of this article may to solve. Only update for the practical value of the association rules, is not only saves updating cost, but also to meet the actual needs of users. A dependency relationship between the antecedent and consequent contains in the association rules, this depends on the strength of the relationship is measured by the size of the confidence of the association rules, bulit a model for accurately reflect the dependency of association rules, and then construct a effective way to updating association rules, has deeply oretical significance and practical value.The main work and contributions can be summarized as follows:· In order to build a Bayesian network for reflect the dependencies of association rules between antecedent and consequent, this thesis take advantage of the dependencies of association rules between antecedent and consequent, given a structure construction algorithm of Bayesian network with conditional dependencies. Based on the directed acyclic graph structure of Bayesian network and the analysis of historical transaction data in the database, this thesis using a likelihood estimation algorithm to calculate the conditional probability tables of the Bayesian network node, then bulit the Bayesian network what we want. · Probabilistic Reasoning is the core of the problem in computing tasks for Bayesian network applied to the actual, consider there may have constructed many of Bayesian Network nodes, in order to ensure the efficiency of reasoning, based on an approximate sampling algorithm, proposed an approximate inference algorithm of Bayesian network, this approximate inference algorithms can predict antecedent attribute state when consequent attribute state has been given, and optimal state of anticedent replace of the incorrected state to complete the association rules update.· Using actual transaction data collected from Internet,this thesis implemente and test the approach of Bayesian network construction and reasoning, the rise in confidence of updated association rules has tested the effectiveness of the method this thesis proposed.