Closed and maximal tree mining using natural representations
José Balcázar, Albert Bifet and Antoni Lozano
In: 5th International Workshop on Mining and Learning with Graphs (MLG'07), 1-3 August 2007, Firenze, Italy.
Mining frequent trees is becoming an important task, with broad applications including chemical informatics, computer vision, text retrieval, bioinformatics, and Web analysis. Many link-based structures may be studied formally by means of unordered trees.
Our focus is on induced rooted trees, thus our relevant information is the root and the link structure. The appropriate notion of subtree, so-called top-down subtree, preserves root and links. We present a new algorithm for mining frequent closed and maximal, ordered
and unordered, labeled trees.