A Neural-Symbolic Approach to Extract Trust Patterns in IoT Scenarios

Trust and reputation relationships among objects represent key aspects of smart IoT object communities with social characteristics. In this context, several trustworthiness models have been presented in the literature that could be applied to IoT scenarios; however, most of these approaches use scalar measures to represent different dimensions of trust, which are then integrated into a single global trustworthiness value. Nevertheless, this scalar approach within the IoT context holds a few limitations that emphasize the need for models that can capture complex trust relationships beyond vector-based representations. To overcome these limitations, we already proposed a novel trust model where the trust perceived by one object with respect to another is represented by a directed, weighted graph. In this model, called T-pattern, the vertices represent individual trust dimensions, and the arcs capture the relationships between these dimensions. This model allows the IoT community to represent scenarios where an object may lack direct knowledge of a particular trust dimension, such as reliability, but can infer it from another dimension, like honesty. The proposed model can represent trust structures of the type described, where multiple trust dimensions are interdependent. This work represents a further contribution by presenting the first real implementation of the T-pattern model, where a neural-symbolic approach has been adopted as inference engine. We performed experiments that demonstrate the capability in inferring trust of both the T-pattern and this specific implementation.