Intelligent Management Of RDF Graph Data Based On Reinforcement Learning

With the continuous development of symbolism,the knowledge has become an important cornerstone of artificial intelligence.In recent years,it has developed rapidly and has many applications.The rise of various applications has led to the increasing scale of knowledge.Some knowledges even have millions of vertices and hundreds of millions of edges,which brings severe challenges to the data management of knowledge.How to effectively store the graph data,index recommendation and query optimization have always been hot and difficult problems of people's research.At present,the fixed-mode storage structure is a major bottleneck of current storage methods.These methods cannot reflect the complexity of graph da ta and adapt to dynamically changing graph datasets and workloads.Secondly,the index recommendation problem is mostly focused on single index recommendation,which undoubtedly limits the efficiency of index recommendation.In terms of query optimization,different connection orders will also lead to different query times.How to find the optimal connection order is also a difficult point in query optimization.In response to the above problems,this paper proposes RDF graph data management technology based on reinforcement learning,and verifies the effects of the proposed method through experiments.Firstly,this paper adopts a relation-based solution for graph storage,uses reinforcement learning to make storage structure decision generation,and dynamically adjusts the storage structure based on database feedback to adapt to dynamically changing graph data and workloads.We have carried out storage structure experiments on PostgreSQL and MySQL databases,and compared the performance with the other four current most advanced storage methods.Extensive experiments on various RDF benchmarks demonstrate that our intelligent storage structure generation method is significantly outperforms the state-of-the-art storage strategies.Secondly,this paper studies the problem of index recommendation,and recommends single-attribute index and multi-attribute index based on reinforcement learning methods.This method considers the interaction between indexes and can further improve query efficiency based on the storage structure.This part has been experimentally verified on the PostgreSQL database and compared with the case of no index and full index.Experiments show that our method has good performance.Thirdly,this paper studies the optimization problem of join orde r in query optimization,improves the shortcomings of existing studies,and proposes a new query sentence feature extraction method.This method embodies the structure information of the connection tree in the process of generating the connection sequence,and is verified by experiments.The experimental results show that our method is superior to the existing latest research REJOIN in performance.Finally,we have systematically integrated the three data management technologies to form a complete intelligent management system of graph data,and carried out the verification and analysis of the integration experiment.