Félicitation à la lauréate 2021 du prix de thèse INFORSID Wafa ABDELGHANI

Wafa a été sélectionnée pour sa thèse intitulée « A Multi-Dimensional Trust-Model for Dynamic, Scalable and Resources-efficient Trust-Management in Social Internet of Things » sous la direction de Florence Sèdes et Ikram Amous (IRIT)

Mots-clefs : Trust Management - Internet of Things - Social Networks - Social Internet of Things - Trust-Attacks.

Résumé : The Internet of Things (IoT) is a paradigm that has made everyday objects intelligent by giving them the ability to connect to the Internet, communicate and interact. The integration of the social component in the IoT has given rise to the Social Internet of Things (SIoT), which has overcome various issues such as interoperability, navigability and resource/service discovery. In this type of environment, participants compete to offer a variety of attractive services. Some of them resort to malicious behaviour to propagate poor quality services. They launch so-called Trust-Attacks (TA) and break the basic functionality of the system. Several works in the literature have addressed this problem and have proposed different trust models. Most of them have attempted to adapt and reapply trust models designed for traditional social networks or peer-to-peer networks. Despite the similarities between these types of networks, SIoT ones have specific particularities. In SIoTs, there are different types of entities that collaborate: humans, devices, and services. Devices can have very limited computing and storage capacities, and their number can be as high as a few million. The resulting network is complex and highly dynamic, and the impact of Trust-Attacks can be more compromising. In this work, we propose a Multidimensional, Dynamic, Resources-efficient and Scalable trust model that is specifically designed for SIoT environments. We, first, propose features to describe the behaviour of the three types of nodes involved in SIoT networks and to quantify the degree of trust according to the three resulting Trust-Dimensions. We propose, secondly, an aggregation method based on Supervised Machine-Learning and Deep Learning that allows, on the one hand, to aggregate the proposed features to obtain a trust score allowing to rank the nodes, but also to detect the different types of Trust-Attacks and to counter them. We then propose a hybrid propagation method that allows spreading trust values in the network, while overcoming the drawbacks of centralized and distributed methods. The proposed method ensures scalability and dynamism on the one hand, and minimizes resource consumption (computing and storage), on the other. Experiments applied to synthetic data have enabled us to validate the resilience and performance of the proposed model.