Querying Knowledge Graphs with Natural Languages

With the unprecedented proliferation of knowledge graphs, how to process query evaluation over them becomes increasingly important. On knowledge graphs, queries are typically evaluated with graph pattern matching, i.e., given a pattern query Q and a knowledge graph G, it is to find the set M(Q, G) of matches of Q in G, where matching is defined with subgraph isomorphism. However querying big knowledge graphs brings us challenges: (1) queries are often issued with natural languages, hence can not be evaluated directly; (2) query evaluation is very costly and match results are often difficult to inspect. In light of these, this paper studies the problem of querying knowledge graphs with natural languages. (1) We extend pattern queries by designating a node \(u_o\) as “query focus”, and revise the matching semantic based on the extension. (2) We develop techniques to understand natural language queries, and generate pattern queries with “query focus”. (3) We develop efficient techniques to identify top-k matches of “query focus”. (4) We experimentally verify that our techniques for query understanding perform well, and our query algorithm is able to find diversified top-k matches efficiently.