Scalable Horn-Like Rule Inference of Semantic Data Using MapReduce.

Semantic data analysis tasks benefit much from rule inference, which derives implicit knowledge from explicit information. Recently, available semantic data from the Web, sensor readings, semantic databases and ontologies exploded drastically. However, most of the existing approaches for semantic rule inference are either centralized, which cannot scale out to infer big semantic data; or rule-specific, which hinder their wildly use. In this paper, we propose a scalable approach for Horn-like rule inference of semantic data based on MapReduce, which can evaluate domain-and application-specific rules, and can be easily extended to evaluate RDFS and OWL ter Horst semantic rules. We first introduce a general rule-evaluation mechanism, which translates a Horn-like rule to one or more MapReduce jobs. To improve rule-evaluation performance, two optimization policies job-parallelization and job-reusing are then introduced. Using a large semantic data set generated by the LUBM benchmark, we give a detailed experimental analysis of the scalability and efficiency of our approaches.