RDFS Inference Search Optimization Method Based on Graph Contraction

Currently, most of RDF stores adopt a materialization approach to the search with RDFS inference. While on-demand approach based on backward reasoning has several desirable properties such as small data size, short load time, and flexibility, it is not popular since its performance is rather low in particular on large RDF graphs. To address this problem, we propose an optimization method based on query transformation. As major RDFS inference rules are recursive, we could not apply naive unfolding technique. The main idea is to make a contracted graph, which is equivalent with respect to RDFS inference but compact, and exploit the result of search with inference on it. Preliminary experimental results on standard LUBM benchmark are encouraging. For some patterns, we confirmed two order of magnitude speedup.