Discovery of Useful Patterns from Tree-Structured Documents with Label-Projected Database

Due to its highly flexible tree structure, XML data is used to capture most kinds of data and provides a substrate in which almost any other data structure may be presented. With the continuous growth of XML tree data in electronic environments, the discovery of useful knowledge from them has been a main research area in the information retrieval community. The mostly used approach to this task is to extract frequently occurring subtree patterns from a set of trees. However, because the number of frequent subtrees grows exponentially with the size of trees, a more practical and scalable alternative is required, which is the discovery of maximal frequent subtrees. The maximal frequent subtrees hold all the useful information, though, the number of them is much smaller than that of frequent subtrees. Handling the maximal frequent subtrees is an interesting challenge, and represents the core of this paper. As far as we know, this is one of the first studies to directly discover maximal frequent subtrees without any candidate sets generations as well as eliminating the process of useless subtree pruning. To this end, we define and use a new type of projected database to represent XML tree data efficiently. It significantly improves the entire process of mining maximal frequent subtree patterns. We study the performance and the scalability of the proposed approach through experiments based on synthetic datasets.