Mining association rules is an important data mining problem which is to find the interested, hidden and unknown knowledge that can be discovered from large database. In many cases, the users just concern a subset of association rules, so the constraint-based association rules mining is introduced. After analyzing and studying constraint-based data mining algorithms, we found there are two problems existing in them:1. Need scan database frequently2. Produce large candidate setsSo it's low efficiency when the algorithms are used to mine low support threshold long-patterns. To solve these problems we introduce an algorithm Con-H-Mine (Constraint-based Hyper-Structure Mining) which is based on H-Mine algorithm and produce no candidate sets. Our algorithm uses Con-H-Struct ( Constraint-based Hyper-Structure) to store transactions so it can reduce the space overhead. Also our algorithm can be fit for large database, it can divide the database accurately.A new good and efficient constraint-based algorithm can be worked out by the above improvement. It realized only providing the interested patterns and improve associative mining. We also design a proto-system of personalize Web site which use this new algorithm, and validate it's actualize.
|