Information-complete and redundancy-free keyword search over large data graphs

Abstract

Keyword search over graphs has a wide array of applications in querying structured, semi-structured and unstructured data. Existing models typically use minimal trees or bounded subgraphs as query answers. While such models emphasize relevancy, they would suffer from incompleteness of information and redundancy among answers, making it difficult for users to effectively explore query answers. To overcome these drawbacks, we propose a novel cluster-based model, where query answers are relevancy-connected clusters. A cluster is a subgraph induced from a maximal set of relevancy-connected nodes. Such clusters are coherent and relevant, yet complete and redundancy free. They can be of arbitrary shape in contrast to the sphere-shaped bounded subgraphs in existing models. We also propose an efficient search algorithm and a corresponding graph index for large, disk-resident data graphs.