A machine learning technique called Graph-Based Induction (GBI) efficiently extracts typical patterns from a directed graph data by stepwise pair expansion (pairwise chunking). We expand the capability of the GBI so that it can handle not only a tree structured data but also a graph data with multi-inputs/outputs nodes and loop structure (including a self-loop) which cannot be treated i nthe conventional way. We show the effectiveness of our approach by applying to the real scale World Wide Web browsing history data.
展开▼
