BPA: A Bitmap-Prefix-tree Array data structure for frequent closed pattern mining

Abstract

This paper presents a new efficient data structure, called "a BPA (Bitmap-Prefix-tree Array)" for discovering frequent closed itemset in large transaction database. Recently, most studies have been focused on using an efficient data structure with preprocessing data for the frequent closed itemset mining. Existing prefix-tree-based approach presented the IT-Tree data structure in its complete preprocessing data for the efficient frequent searching but used large memory space and time consuming in the preprocessing step.Lately, another approach introduced the efficient data structure, called "a collaboration of array, bitmap, and prefix tree", to improve storage and time in preprocessing data. However, its preprocessing step was not complete and hence its frequent searching for the frequent closed itemset mining may take more time than that of the IT-Tree-based approach. In this paper, we propose the efficient BPA data structure to enhance not only computation-time and memory-space in the complete preprocessing data but also in those in the frequent searching.

EBPA: An efficient data structure for frequent closed itemset mining

“…We design the efficient EBPA data structure as an array-based (compact) prefix tree for saving space (in practice) that concerns only occurrence transactions and their corresponding parent nodes, an improved version of our previous data structure (BPA) [14] that exponentially generates full nodes (N) of the complete prefix tree (Fig.3b), where N = 2 0 + 2 1 + 2 2 + … + 2 i + … + 2 m-1 = 2 m -1 nodes. In particular, there are two main steps in the collaboration based EBPA-FCIM mining: 1) create the (compact) prefix tree ( Fig.4b) and the (bucket) arrays (Fig.4c) in Section 3.2 and 2) compute the closure sets and post sets of the FCIM mining (see Section 3.3).…”

“…Later research focuses on the complete closed itemsets, which can reduce a number of itemsets without information loss and can represent covering all results of the original FIM with saving memory space and time. Therefore, many FCIM approaches have been proposed [1] - [14].…”

“…Besides the best solution, an appropriate data structure and corresponding frequency computation functions are a key to improve the performance of each method. Recently, many data structures for computing the FCIM have been proposed ( [4], [7], [10], [12], [14]).…”

“…In this paper, we propose the improved collaboration data structure, called the EBPA (Efficient Bitmap-Prefix-tree Array) and the EBPA-CLOSED algorithm for frequent closed itemset mining. The EBPA data structure, the (top down) prefix-tree construction, is an improved version of our previous work, the (bottom up) prefix-tree BPA [14]. In order to gain the advantage of the collaboration approach, our EBPA maintains all features and results of the existing collaboration data structure [10] without extra sorting and merging transactions.…”