Towards efficient processing of graph queries

Abstract

Conceptually, a graph consists of a set of vertices and edges where both vertices and edges may have attributes to represent complex information associated with themselves. There are a lot of real-world datasets, which can be easily expressed as graphs. For example, it is natural to represent computer networks, food webs, protein interactions, and social networks as graphs. From the users` point of view, the main concern is how to quickly find information they needed from these graph datasets after issuing graph-structured queries. The goal of this thesis is to provide efficient ways of processing graph queries over large graph datasets.

In this thesis, we focus on two types of graph query problems, i.e., subgraph matching and graph containment. We first answer to a subgraph matching query problem. The results of a subgraph matching query are all occurrences of the query in a target data graph. Since a target data graph typically has a lot of vertices, the main challenge of the problem is how to filter out unqualified vertices, which are not eligible to be mapped to vertices in the query. Here we analyze all possible vertex features, which are made of the combination of label distribution and connectivity information within two hops of neighbor of a vertex, and choose vertex features that are balanced with pruning power and extraction cost. Then, we propose a feature indexing method over the chosen vertex features, which is used for finding candidates for each vertex in query processing.

Meanwhile, due to potential errors and noises in real-world graph datasets, exact subgraph matching may be sometimes inappropriate in practice. Thus approximate subgraph matching is also required. In this thesis, we investigate an approximate subgraph matching model that allows missing edges. A simple solution is to first generate all query subgraphs from a query within a threshold and perform exact subgraph matching for each query subgraph separately. The main challenge o...