MRQUSAR: A web-scale distributed spatial reasoner using MapReduce

In order to answer effectively on behalf of humans in a DeepQA environment, such as the American quiz show Jeopardy (http://www.jeopardy.com), the computer is required to have the capability of fast temporal and spatial reasoning on a large-scale commonsense knowledge base. Many existing spatial reasoners share a common limitation in that they do not contain conversion rules between the directional and topological relationship set. Furthermore, they are limited in scalability because their algorithms are designed for centralized processing. In this paper, we propose an efficient scalable spatial reasoning algorithm that includes conversion rules to manage directional and topological relationships, and then explain the implementation of a scalable spatial reasoner based on the proposed algorithm using the MapReduce framework. Our algorithm is designed for distributed parallel processing, and thus the scalability of the algorithm can be enhanced in two dimensions. This paper also presents the principle of a spatial knowledge generator developed to build a large-scale spatial knowledge base, and then introduces the results of the experiment conducted to measure the performance of our algorithm.