Query By Example On Multiple Knowledge Graphs

With the explosive growth of knowledge and the blooming of knowledge graph in different domain,querying on knowledge graph has become an attractive topic in the domain of search engine.However,due to the independence of knowledge graphs in different domains,it would not meet users' requirements only querying on a single knowledge graph.In this thesis,we propose a novel method to query by example on multiple knowledge graphs,which ensures the efficiency of time,at the same time,it improves the quality of the results and the users' satisfaction by improving the method of candidate results fusing and perfecting results relevance calculation.Most of the existing research is querying on a single knowledge graph.And the existing technology of querying on graph can't be applied on querying on multiple graphs.At the same time,the existing method of results relevance calculation is based on traditional data graph,without considering the difference of knowledge and data.To resolve the problems above,three parts of research are conducted in this thesis:The first part is about an model for the querying by example on the multiple graphs.This system use user-friendly keyword query technology.Firstly,it structure user's input of query keywords as a query sample.Then,after determining the position of the query sample on each knowledge graph,it use the method of subgraph isomorphism to find k high relevant sub-graphs according to the query sample on each knowledge graph.Finally,candidate sub-results have isomorphisming with users' query sample fuse together.Algorithm integrates different knowledge graphs by integrating Top-K results from different knowledge graphs,avoiding the global schema integration of knowledge graphs,and more flexibility.Experiments show that the method can ensure the query efficiency and improve the quality of the query results.The second part is about the method of query results relevance calculation based on knowledge.In order to further capture the user's query intention,this thesis adds the prevalence of knowledge as a new factor to supplement the method of results relevance calculation based on the existing distance and structure.This thesis regard the event happened time as a measure of prevalence of knowledge,because the closer that happened,the more popular knowledge.The experiments have shown that the improved methods can effectively improve the quality of query results and users' satisfaction.The third part is on the method of candidate results fusion algorithm based grouping and labeling.In the query by example on multiple knowledge graphs,the query results may be from a single knowledge graph or fusing of candidate sub-results from different knowledge graphs.To solve the high cost of the candidate results fusion due to the too many candidate results,this thesis propose an optimization algorithm.It groups and labels for the candidate results according to the characteristics,and fuses the candidate according to the set.So it can reduce the cost of node matching,shorten the response time.The experiments in this thesis verify that the method can effectively improve the query efficiency.