The Web of Things extends the Internet of Things with standards of the Web, such as knowledge graphs and reasoning abilities. Traditional reasoners are optimized for performance and require excessive hardware resources, especially memory, to run on the constrained objects typically used in IoT applications. In this thesis, we present optimizations of a commonly used reasoning algorithm to make its implementation possible on constrained objects, with a focus on reducing the required memory footprint for reasoning. These optimizations are (i) algorithmic, taking advantage of the characteristics of the rulesets used for reasoning, and (ii) focused on in-memory knowledge representation within the reasoner. We also present LiRoT, the implementation of a lightweight, incremental reasoner that can be embedded on constrained objects, as a result of these optimizations. We experimentally evaluate the impact of the various optimizations, and show that they significantly reduce the memory footprint of the reasoner, with minimal impact on computation time. We also compare LiRoT to several state of the art reasoners, and show that it is better suited for handling datasets with sizes typical of IoT applications. Finally, we evaluate the use of LiRoT on platforms such as ESP32 and Arduino Due, and show that it is possible to perform reasoning tasks on such constrained platforms. This thesis was funded by the CoSWoT project (ANR-19-CE23-0012).

The main objectives of this PhD is to provide contributions to embedded reasoning on the Web of Things.