Query Processing on Attributed Graphs

An attributed graph is a powerful tool for modeling a variety of information networks. It is not only able to represent relationships between objects easily, but it also allows every vertex and edge to have its attributes. Hence, a lot of data, such as the web, sensor networks, biological networks, economic graphs, and social networks, are modeled as attributed graphs. Due to the popularity of attributed graphs, the study of attributed graphs has caught attentions of researchers. For example, there are studies of attributed graph OLAP, query engine, clustering, summary, constrained pattern matching query, and graph visualization, etc. However, to the best of our knowledge, the studies of topological and attribute relationships between vertices on attributed graphs have not drawn much attentions of researchers. Given the high expressive power and popularity of attributed graph, in this thesis, we define and study the processing of three new attributed graph queries, which would help users to understand the topological and attribute relationships between entities in attributed graphs. For example, a reachability query on a social network can tell whether two persons can be connected given certain attribute constraints; a reachability query on a biological network can tell whether a compound can be transformed to another compound under given chemical reaction conditions; a How-to-Reach query can tell why the answers of the above two reachability query are negative; a visualizable path summary query can offer an overall picture of topological and attribute relationship between any two vertices in attributed graphs. Except for the proposed query types in this thesis, we believe that there is still penalty of meaningful attributed graph query types that have not been proposed and studied by the database and data mining community since an attributed graph is a very rich source of information. Through this thesis, we hope to draw people's attentions on attributed graph query processing so that more hidden information contained in attributed graphs can be queried and discovered.