CLAIM: An Efficient Method for Relaxed Frequent Closed Itemsets Mining over Stream Data

Abstract

Recently, frequent itemsets mining over data streams attracted much attention. However, mining closed itemsets from data stream has not been well addressed. The main difficulty lies in its high complexity of maintenance aroused by the $\underline{exact}$ model definition of closed itemsets and the dynamic changing of data streams. In data stream scenario, it is sufficient to mining only approximated frequent closed itemsets instead of in full precision. Such a compact but close-enough frequent itemset is called a relaxed frequent closed itemsets.

In this paper, we first introduce the concept of $\mathcal{RC}$ (Relaxed frequent Closed Itemsets), which is the generalized form of approximation. We also propose a novel mechanism CLAIM, which stands for CLosed Approximated Itemset Mining, to support efficiently mining of $\mathcal{RC}$. The CLAIM adopts bipartite graph model to store frequent closed itemsets, use Bloom filter based hash function to speed up the update of drifted itemsets, and build a compact HR-tree structure to efficiently maintain the $\mathcal{RC}$s and support mining process. An experimental study is conducted, and the results demonstrate the effectiveness and efficiency of our approach at handling frequent closed itemsets mining for data stream.

This work is supported by the National Natural Science Foundation of China under Grant No. 60473051 and No.60642004 and HP and IBM Joint Research Project.