Efficiently Mining Homomorphic Patterns from Large Data Trees

Finding interesting tree patterns hidden in large datasets is an central topic in data mining with many practical applications. Unfortunately, previous contributions have focused almost exclusively on mining induced patterns from a set of small trees. The problem of mining homomorphic patterns from a large data tree has been neglected. This is mainly due to the challenging unbounded redundancy that homomorphic tree patterns can display. However, mining homomorphic patterns allows for discovering large patterns which cannot be extracted when mining induced or embedded patterns. Large patterns better characterize big trees which are important for many modern applications in particular with the explosion of big data. In this paper, we address the problem of mining frequent homomorphic tree patterns from a single large tree. We propose a novel approach that extracts nonredundant maximal homomorphic patterns. Our approach employs an incremental frequency computation method that avoids the costly enumeration of all pattern matchings required by previous approaches. Matching information of already computed patterns is materialized as bitmaps a technique that not only minimizes the memory consumption but also the CPU time. We conduct detailed experiments to test the performance and scalability of our approach. The experimental evaluation shows that our approach mines larger patterns and extracts maximal homomorphic patterns from real datasets outperforming state-of-the-art embedded tree mining algorithms applied to a large data tree.