An Efficient Graph Mining System for Large Patterns.
Peng Ming Jiang, Rujia Wang, Bo Wu · arXiv (Cornell University) · 2021
There is a growing interest in designing systems for graph pattern mining in recent years. The existing systems mostly focus on small patterns and have difficulty in mining larger patterns. In this work, we propose Angelica, a single-machine graph pattern mining system aiming at supporting large patterns. We first propose a new computation model called multi-vertex exploration. The model allows us to divide a large pattern mining task into smaller matching tasks. Different from the existing systems which perform vertex-by-vertex exploration, we explore larger subgraphs by joining small subgraphs. Based on the new computation model, we further enhance the performance through an index-based quick pattern technique that addresses the issue of expensive isomorphism check, and approximate join that mitigates the issue of subgraph explosion of large patterns. The experimental results show that Angelica achieves significant speedups against the state-of-the-art graph pattern mining systems and supports large pattern mining that none of the existing systems can handle.