Towards Hybrid Methods for Graph Pattern Queries

In the subgraph querying problem, given a query graph and a dataset of graphs, we want to locate which graphs in the dataset contain the query and/or find all its occurrences. Over the years, numerous methods, fragmented in 2 categories, were proposed to tackle the problem. In the first category, methods follow the filter-then-verify (FTV) paradigm where an index is used to filter out graphs that definitely do not contain the query as an answer. On the remaining set of graphs, a subgraph isomorphism algorithm is applied to verify the graphs that contain the query graph. A second category, so called no-filter, verify (NFV), invested in optimizing the subgraph isomorphism process, by employing a lightweight index primarily to locate candidate vertices on the graphs in the dataset. The current trend is to totally dismiss the FTV methods and employ NFV methods instead. With this work we wish to point out that a hybrid solution that is, a combination of the filtering of some FTV method with the subgraph isomorphism test of top-performing NFV methods, can be highly competitive in terms of performance vis-a-vis methods from either the FTV or the NFV category. We will present some preliminary results, based on experiments, where such a combination proves to be beneficial by primarily avoiding the initiation of a large number of redundant subgraph isomorphism tests.