A Query Decomposition Method For Large Scale Knowledge Graph

In recent years,with the increasing number of Internet users,the scale of user data is also growing.In such a large amount of data,knowledge graph data is of vital importance.Data in many fields can be represented by the knowledge graph data model,such as social network,knowledge graph,and recommendation algorithm.Large-scale knowledge graph data processing usually adopts a distributed solution.Distributed graph databases need to address two key points: data distribution information and query decomposition.In order to provide the necessary index information for query decomposition,a bidirectional hash partition based on subject and object is used.On that foundation,the range index is applied to record the range value of variables,and the storage structure of it is designed.When importing data,the range of variables is recorded and conserved,so it can provide information input while query rewriting during operation.Multiple binary search algorithm is used to improve the retrieval efficiency.In terms of query decomposition,the algorithm of star query decomposition based on the vertex degree is adopted in query decomposition,and the query is optimized by query rewriting and cost estimation.The algorithm divides the original query into several local subqueries according to the degree information of the graph vertex,then rewrites the local subquery with the information provided by the range index to constrain the number of intermediate result sets and reduces the network communication overhead.The experiment is divided into two parts for testing.As far as data partitioning,our algorithm is tested by using a variety of datasets of different sizes.The results show that our algorithm has linear data partitioning time and a stable data redundancy rate.With respect to query decomposition,the test compared the performance data before and after the optimization and analyzed it with other algorithms.It turns out that the query time after the optimization of our algorithm is 30% faster than that before the optimization,and its query time is 15% faster than the contrastive algorithm.