Social Internet of Things (SIoT) is a paradigm in which the Internet of Things (IoT) concept is fused with Social Networks for allowing both people and objects to interact in order to offer a variety of attractive services and applications. However, with this emerging paradigm, people feel wary and cautious. They worry about revealing their data and violating their privacy. Without trustworthy mechanisms to guarantee the reliability of user’s communications and interactions, the SIoT will not reach enough popularity to be considered as a cutting-edge technology. Accordingly, trust management becomes a major challenge to provide qualified services and improved security. Several works in the literature have dealed with this problem and have proposed different trust-models. Nevertheless, proposed models aim to rank the best nodes in the SIoT network. This does not allow to detect different types of attack or malicious nodes. Hence, we overcome these issues through proposing a new trust evaluation model, able to detect malicious nodes, block and isolate them, in order to obtain a reliable and resilient system. For this, we propose new features to describe and quantify the different behaviors that operate in such system. We formalized and implemented a new function learned and built based on supervised learning, to analyze different features and distinguish malicious behavior from benign ones. Experimentation made on a real data set prove the resilience and the performance of our trust model.