Praedixi, Redegi, Cogitavi: Adaptive knowledge for resource-aware semantic reasoning

Abstract

Representing knowledge with ontologies and performing reasoning with semantic reasoners is important in many intelligent applications. However, existing reasoners do not take into account the available resources of the device where they run, which can be important in many scenarios such as reasoning with very large ontologies or reasoning on resource-constrained mobile devices.In this paper, we propose a novel approach to adapt the size of knowledge managed by applications, taking into account several criteria about resources available (such as time, memory, and battery consumption), at the same time. Thus, rather than giving no answer due to the lack of resources needed to deal with a full ontology, we propose a novel architecture to compute a subontology to provide an incomplete answer at least. Our approach makes use of existing approaches to predict the performance of semantic reasoners and to compute ontology modularisation and ontology partition, but taking into account the associated resource consumption. We also propose a novel measure to estimate the semantic loss when replacing the original ontology by a subontology. Finally, we present an implementation and evaluation of the whole pipeline, showing that the semantic loss incurred in the process is acceptable.