Research And Implementation Of Algorithms Of Mining Association Rules Based On Concept Lattice

The quick development of information technology leads an incredible rocket in all kinds of data. Large, distributed databases have been more and more popular. Just these cases bring birth to the KDD and Data Mining providing people a new approach to understand data. Due to a complete structure and solid mathematics theory, Concept Lattice has been accepted as the natural model of data mining.As an important pattern in data mining, association rules always attract many researchers. In this dissertation a new method is presented to mine association rules based on the model of concept lattice. By effectively building concept lattice, this method generates association rules after finding frequent itemsets. Taking into account that many current popular algorithms finding frequent itemsets not only need to scan databases lots of times, but also lead large amounts of unnecessary costs when dealing with those databases maintained from time to time. Our method only scans databases once. It can generate firstly the largest closed pattern and then get all frequent itemsets with good performance. What is more vital is that the concept lattice built during the mining process can be maintained easilier when databases change. Meanwhile, the concept lattice can be stored to mine other patterns.This dissertation distusses the principles of building concept lattice and reduced extended concept lattice, presents new algorithms to build them, and describes two methods to prune concept lattice for mining association rules. Based on these, the algorithm of finding association rules and the results of experiments are also shown in detail in this dissertation. At last, one kind of architecture of distributed mining association rules is presented. This architecture can also be used to mine very large databases and it is proved correct by experiments.