KR Conference 2012 Short Paper
- Ilias Tachmazidis
- Grigoris Antoniou
- Giorgos Flouris
- Spyros Kotoulas
which encode commonsense, practical knowledge that humans possess and would allow the system to automatically reach useful conclusions based on the provided data and infer new and useful knowledge based on the data. For example, in ((Urbani et al. 2009)) for 78, 8 million statements crawled from the Web, the number of inferred conclusions (RDFS closure) consists of 1, 5 billion triples. In this paper, we consider nonmonotonic rule sets ((Antoniou and van Harmelen 2008), (Maluszynski and Szalas 2010)). Such rule sets provide additional benefits because they are more suitable for encoding commonsense knowledge and reasoning. In addition, nonmonotonic rules avoid triviality of inference, which could easily occur when lowquality raw data is fed to the system; the latter is common in this setting, given the interconnection of data from different sources, over which the data engineer has no control. The main challenge rising in such a setting is the feasibility of reasoning over such large volumes of data. One of the most promising methods to address this problem is by using massively parallel reasoning processes that would handle reasoning by using several computers in the cloud, assigning each of them a part of the parallel computation. In the last two of years, there has been significant progress in parallel reasoning e. g., in ((Oren et al. 2009)), ((Urbani et al. 2009)), ((Kotoulas, Oren, and van Harmelen 2010)), ((Goodman et al. 2011))), scaling reasoning up to 100 billion triples ((Urbani et al. 2010)). Nevertheless, current approaches have been restricted to monotonic reasoning, namely RDFS and OWL-horst, or have not been evaluated for scalability ((Mutharaju, Maier, and Hitzler 2010)). However, in many application scenarios, one needs to deal with poor quality data (e. g., involving inconsistency or incompleteness), which could easily lead to reasoning triviality when considering rules based on monotonic formalisms; this problem can be managed with nonmonotonic rules and nonmonotonic reasoning. We study the problem of reasoning over huge datasets equipped with nonmonotonic (defeasible) rules using massively parallel (cloud) computational techniques. Following previous works, we adopt the MapReduce framework ((Dean and Ghemawat 2004)), suited for parallel processing of huge datasets. A restricted form of defeasible logic is studied: singleargument defeasible logic. A MapReduce algorithm is pre- We are witnessing an explosion of available data from the Web, government authorities, scientific databases, sensors and more. Such datasets could benefit from the introduction of rule sets encoding commonly accepted rules or facts, application- or domain-specific rules, commonsense knowledge etc. This raises the question of whether, how, and to what extent knowledge representation methods are capable of handling the vast amounts of data for these applications. In this paper, we consider nonmonotonic reasoning, which has traditionally focused on rich knowledge structures. In particular, we consider defeasible logic, and analyze how parallelization, using the MapReduce framework, can be used to reason with defeasible rules over huge data sets. Our experimental results demonstrate that defeasible reasoning with billions of data is performant, and has the potential to scale to trillions of facts. 1 Grigoris Antoniou Institute of Computer Science, FORTH University of Huddersfield, UK antoniou@ics. forth. gr