Research On Semantic Similarity Graph Query Over Knowledge Graphs

Knowledge graph models objective facts in real world and represents knowledge in the form of graphs.In knowledge graph,entities can be viewed as nodes and the relationships can be denoted as edges.In recent years,many well-known knowledge graphs have been built,such as YAGO,DBpedia and Freebase.How to effectively query on knowledge graph to obtain relevant information is a current research hotspot.The research results have important research significance,which can be applied to intelligent question answering system,recommender system,etc.Graph query technology is an important query method in knowledge graph query.How to achieve an efficient graph query method is the key to effective knowledge graph query.Existing graph query algorithms usually perform subgraph matching based on structural similarity.However,they ignore the rich semantic information in the knowledge graph,which will return incomplete query results and affect query accuracy.Moreover,most graph queries are deployed on entire knowledge graphs,which will affect query efficiency,especially in cases where the knowledge graph is large.This paper mainly provides the following research work to solve the above problems:(1)Topic subgraph mining algorithm.First,the original knowledge graph data is preprocessed to eliminate noise data.Then,a topic subgraph mining algorithm is proposed,which applies type similarity to divide the knowledge graph into multiple topic graphs.The topic graphs are independent,which is beneficial to achieve distributed storage and improve query efficiency.(2)Knowledge graph storage architecture based on topic graph.Based on adjacency table and skip list index,a reasonable storage architecture is designed to store topic graphs.Then several APIs(insert,delete,update,and query)are provided to achieve effective management of topic graphs.They can provide reliable underlying data support for semantic similarity graph queries.(3)Graph query algorithm and optimization based on semantic similarity.First,the graph embedding model is used to represent predicates into multi-dimensional vector space.Based on the semantic similarity of predicates,the dynamic semantic graph is constructed.Then,a graph query algorithm based on semantic similarity is provied to return the top-k results by modifying the heuristic function of A*algorithm.It cat prune search paths by the upper/lower bounds of the path semantics.Finally,the time-constrained query optimization is performed on the semantic similarity query algorithm,which can return approximate results within the user's time constraints by iterating the early-explored results and estimating the query time.Finally,based on the above research results,a knowledge graph semantic similarity query system is designed.Extensive experiments on query algorithm confirm the effectiveness and efficiency of this system.