Algorithm for Enumerating All Maximal Frequent Tree Patterns among Words in Tree-Structured Documents and Its Application

To extract structural features from tree-structured documents among nodes in which characteristic words appear, we described a text-mining algorithm for enumerating all frequent consecutive path patterns (CPP) on a list W of words in Uchida et al., PAKDD 2004 [14]. In this paper, we first extend a CPP to a tree pattern, which is called a tree association pattern (TAP), over a set W of words. A TAP is an ordered rooted tree t such that the root of t has no child or at least two children, all leaves of t are labeled with non-empty subsets of W and all internal nodes, if they exist, are labeled with strings. By modifying text-mining algorithms to find all frequent CPPs, next, we present text-mining algorithms for enumerating all maximal frequent TAPs in tree-structured documents, where a TAP t is maximal if there exists no frequent TAP, which has t as a proper subpattern. Then, we tested our algorithms using Reuters news wires. Finally, as one application of CPPs, we present an algorithm for a wrapper based on CPP using XSLT transformation language and demonstrate simply the use of the wrapper to translate one of the Reuters news wires into other XML document.