Discovering frequent substructures in large unordered trees

Abstract

In this paper, we study a frequent substructure discovery problem in semi-structured data. We present an efficient algorithm Unot that computes all frequent labeled unordered trees appearing in a large collection of data trees with frequency above a user-specified threshold. The keys of the algorithm are efficient enumeration of all unordered trees in canonical form and incremental computation of their occurrences. We then show that Unot discovers each frequent pattern T in O(kb2m) per pattern, where k is the size of T, b is the branching factor of the data trees, and m is the total number of occurrences of T in the data trees. © Springer-Verlag Berlin Heidelberg 2003.