Efficient Substructure Discovery from Large Semi-structured Data

By rapid progress of network and storage technologies, a huge amount of electronic data such as Web pages and XML data [23] has been available on intra and internet. These electronic data are heterogeneous collection of ill-structured data that have no rigid structures, and often called semi-structured data [1]. Hence, there have been increasing demands for automatic methods for extracting useful information, particularly, for discovering rules or patterns from large collections of semi-structured data, namely, semi-structured data mining [6, 11, 18, 19, 21, 25]. In this paper, we model such semi-structured data and patterns by labeled ordered trees, and study the problem of discovering all frequent tree-like patterns that have at least a minsup support in a given collection of semi-structured data. We present an efficient pattern mining algorithm FREQT for discovering all frequent tree patterns from a large collection of labeled ordered trees. Previous algorithms for finding tree-like patterns basically adopted a straightforward generate-and-test strategy [19, 24]. In contrast, our algorithm FREQT is an incremental algorithm that simultaneously constructs the set of frequent patterns and their occurrences level by level. For the purpose, we devise an efficient enumeration technique for ordered trees by generalizing the itemset enumeration tree by Bayardo [10]. The key of our method is the notion of the rightmost expansion, a technique to grow a tree by attaching new nodes only on the rightmost branch of the tree. Furthermore, we show that it is sufficient to maintain only the occurrences of the