Identifying User Intent and Context in Graph Queries
share
Graph querying is the task of finding similar embeddings of a given query graph in a large target graph. Existing techniques employ the use of structural as well as node and edge label similarities to find matches of a query in the target graph. However, these techniques have ignored the presence of context (usually manifested in the form of node/edge label distributions) in the query. In this paper, we propose CGQ (Contextual Graph Querying), a context-aware subgraph matching technique for querying real-world graphs. We introduce a novel sub-graph searching paradigm, which involves learning the context prevalent in the query graph. Under the proposed paradigm, we formulate the most contextually-similar subgraph querying problem that, given a query graph and a target graph, aims to identify the (top-k) maximal common subgraph(s) between the query and the target graphs with the highest contextual similarity. The quality of a match is quantized using our proposed contextual similarity function. We prove that the problem is NP-hard and also hard to approximate. Therefore, to efficiently solve the problem, we propose a hierarchical index, CGQ-Tree, with its associated search algorithm. Our experiments show that the proposed CGQ index is effective, efficient and scalable.
READ FULL TEXT