Frequent pattern mining in attributed trees: algorithms and applications

Abstract

Frequent pattern mining is an important data mining task with a broad range of applications. Initially focused on the discovery of frequent itemsets, studies were extended to mine structural forms like sequences, trees or graphs. In this paper, we introduce a new domain of patterns, attributed trees (atrees), and a method to extract these patterns in a forest of atrees. Attributed trees are trees in which vertices are associated with itemsets. Mining this type of patterns (called asubtrees), which combines tree mining and itemset mining, requires the exploration of a huge search space. To make our approach scalable, we investigate the mining of condensed representations. For attributed trees, the classical concept of closure involves both itemset closure and structural closure. We present three algorithms for mining all patterns, closed patterns w.r.t. itemsets (content) and/or structure in attributed trees. We show that, for low support values, mining content-closed attributed trees is a good compromise between non-redundancy of solutions and execution time.