Similarity Search in Large-Scale Graph Databases

Peixiang Zhao

Abstract: Graphs are ubiquitous and play an essential role in modeling and representing complex structures in real-world networked applications. Given a graph database that comprises a large collection of graphs, it is fundamental and critical to enable fast and flexible search for structurally similar graphs. In this paper, we survey recent graph similarity search techniques and specifically focus on the work based on the graph edit distance (GED) metric. State-of-the-art approaches for the GED based similarity search typically adopt a pruning and verification framework. They first take advantage of some easy-to-compute lower-bounds of graph edit distance, and use novel graph indexing structures to efficiently evaluate such lower-bounds between graphs in the graph database and the query graph. This way, graphs that violate the GED lower-bound constraints can be identified and filtered from the graph database from further investigation. Then, the costly GED verification is performed only for the graphs that pass the GED lower-bound evaluation. We examine existing GED lower-bounds, graph index structures, and similarity search algorithms in detail, and compare different similarity search methods from multiple aspects including index construction cost, similarity search performance, and applicability in real-world graph databases. In the end, we envision and discuss the future research directions related to similarity search and high-performance query processing in large-scale graph databases.