A Lattice-based Graph Index for Subgraph Search

Given a query graph q, a “subgraph-search” algorithm retrieves from a graph database D all graphs that have q as a subgraph, D(q). Subgraph search is costly because of its involvement of a subgraph-isomorphism test, which is a NPcomplete problem. Graph indexes are used to improve the algorithm efficiency by first filtering out a set of false answers and then verifying each graph that has passed the filtration with subgraph isomorphism tests. Many substructure features have been proposed to build the index aiming at improving the filtering power of the index. In this paper we improve the filtering power and query processing time by design of the index structure. We propose a lattice like index, Lindex, which is generally applicable on all graph features. Lindex achieves a high filtering rate by organizing index subgraphs in a graph lattice and adopting a specific design of value sets. Besides finding the candidate set C(q) after filtering, Lindex can also find a set of true answers Tr(q) without involving subgraph isomorphism tests. Accordingly, only candidate graphs in C(q) − Tr(q) need to be verified. Our experiments show that Lindex outperforms other cutting-edge indexes on both frequent subgraph and infrequent subgraph queries.