Result Summarization for Graph Search

[VLDB'14 Paper]

In real search applications, a query on big graphs might produce an excessive number of results (a.k.a. "answer graphs" in graph search) which may convey different meanings. This makes the understanding of the results a daunting task for the users. Our work addresses the challenge by proposing a novel notion of "summary graph" which could precisely describe the answer graphs with similar meanings. By summarizing the results into several small summary graphs, users can easily get intuitive "big pictures" of all the answers, as illustrated by the example below.

Graph Search with Ontology

[ICDE'13 Paper][Project homepage]

Many search systems are designed for IR-based matching which could only retrieve answers that are syntactic-close to the query. Such approaches are often too restrictive to provide users with the information that are actually semantic-close to the query. Our work extends the existing graph search to identify semantically related matches by leveraging ontology graph (see an example below). The work proposes a novel ontology-based indexing technique that could help search top-K answers efficiently.

Distributed Graph Management

[SIGMOD'12 Paper][Project homepage]

With the prevalence of big graph data, e.g., Facebook's social network and Google's knowledge graph, there is a practical need for distributed systems that are able to process large graphs efficiently. In this work, we introduce two novel graph partitioning mechanisms, complementary partitioning and dynamic partitioning, with the goal of answering graph queries locally. Moreover, our method could minimize inter-machine communication by dynamically adapting the partitions to query workload change, as well as data change. See the architecture below.