Research On Key Technology Of Parallel Inference On Large-scale RDF Graph

Large volumes of RDF data have been published with the rapid development of the Semantic Web.RDF data of each research area presents a geometric explosive growth.The number of triples has been far more than a scale of ten billion.Large-scale RDF data not only increases the complexity of management,but also poses an enormous challenge to reasoning over semantic data.How to reason on RDF data efficiently becomes the focus of most research work.Most current semantic inference engines are difficult to adapt to the reasoning requirements of large-scale semantic data,since the execution process is time consuming.To solve this problem,this paper proposes an RDFS parallel inference framework based on message passing mechanism.The graph structure of RDF data is exploited to abstract inference process to an “edge addition model” according to the characteristics of RDFS rules.We map the process of acquiring derivations of rules into the process of adding new edges of RDF graph in “edge addition model”.At the same time,to reduce the number of iterations of parallel reasoning process,we analyze the dependencies of RDFS rules to arrange the optimal execution order.The whole calculation procedure is vertex centric.Vertices execute the parallel inference algorithm,which can send messages of inference to other vertices to complete inference process.When all derivations are regarded as new edges of initial RDF graph,the computation terminates.In the end,we design the parallel inference algorithm based on Pregel model which is based on message passing mechanism.MPPIE,the RDFS parallel inference framework,is implemented on top of open source framework Giraph.It has been proved with a lot of experiments that this algorithm is correct and keeps high performance,especially for big data.The average execution time of reasoning is 30 times faster than the state-of-art semantic scalable inference engine.This paper proposes MPPIE,an RDFS parallel inference framework based on message passing mechanism.The analysis and experimental results present significant performance advantages of our approach of RDFS inference based on message passing mechanism.Experimental results on both benchmark and real world datasets show the performance of our method outperforms WebPIE,the state-of-art semantic scalable inference engine.Furthermore,our method provides good scalability.