Research Of Example Query Method On Ontology-labels Knowledge Graph

With the explosive growth of data,knowledge graphs,as a semantic network,are widely used in recommendation systems,knowledge question answering,social network analysis,and other fields.In the era of big data,with the rapid popularization of the Internet,massive data from the internet and real-life has brought valuable data wealth to people but also brought great challenges to data retrieval.Therefore,how to query knowledge graphs with rich semantic information efficiently to obtain interesting results has aroused extensive discussion in academia and industry,which has important research value and significance.In the early days,the query of knowledge graph was mainly based on RDF triple data query.RDF triple was stored in the relational database,and structured query language was used to find the results that met the query conditions.However,it is very difficult for users who are not domain experts to express their query intentions and interests.Graph-based query method makes users get rid of the trouble of learning complex query language,and only needs users to provide graph structure that meets their query conditions.Its core idea is subgraph matching,which finds isomorphic subgraphs as answers in knowledge graphs through graph matching.The example query based on graph structure does not strictly require the characteristics of query conditions and regards the user's query as a data example that users are interested in,which is convenient for users to express their query intentions.Most of the example query methods on traditional knowledge graphs are based on node-labels or edge-labels for subgraph matching.In the knowledge graph with rich semantic information,node-labels can only identify the name and attribute information of entities,and the same edge-labels may connect with many different types of entities.Therefore,the example query based on the traditional knowledge graph will lead to low semantic relevance of query results,and the returned query results can not reflect users' query interests well.To solve the above problems,we propose an example query method for ontology-labels knowledge graph.The main research work and innovations are as follows:(1)To solve the problem of low semantic relevance of traditional knowledge graph query results,we propose to introduce the ontology-labels set,that is,entity type and subtype,as the semantic description of entity nodes.In the process of sub-graph matching,both ontology-labels similarities of entity nodes and edge-labels isomorphism are considered to retrieve answers with higher semantic relevance.(2)A single example query method based on ontology-labels knowledge graph is proposed for users to input a single query graph.The first stage is the candidate node filtering stage.Firstly,the effective bidirectional index and ontology-labels tree index are proposed to reduce the search space in advance and determine the candidate range of query nodes.Secondly,a formula for calculating the similarity of ontology-labels is proposed,which sorts the candidate nodes in descending order according to the correlation score of ontology-labels to form an orderly set of candidate nodes,which is convenient for verifying the candidate nodes with high correlation first.The second stage is the verification stage.Firstly,the edge-label isomorphism algorithm is used to form an ordered set of matching pairs,and then the effective candidate result combination sorting algorithm is used to directly combine the first k answers with the highest correlation.(3)To solve the problem of users inputting multiple query graphs,we propose a multi-example query method on the ontology-labels knowledge graph.Firstly,the ontology-edge label index is used to reduce the search space,avoid searching in the whole knowledge graph,and improve query efficiency.Secondly,the matching pattern is constructed by using the example set of the user query,to give priority to verifying the more compact answers and preparing for the matching verification algorithm.Thirdly,according to the matching pattern,the candidate nodes are verified by the combination of node vector and the index in the reduced search space,and the candidate space is further refined.Finally,the candidate cardinality of matching pattern fragments is evaluated in the candidate space,and the structure with the least candidate cardinality is selected for extended matching to reduce the number of isomorphic matching before the answer set is formed and returned to the user.(4)A large number of experiments on two real data sets show that our proposed algorithms are superior to some existing algorithms in efficiency and effectiveness.