Social Internet of Things (SIoT) is a new paradigm where the Internet of Things (IoT) is merged with social networks, allowing objects to establish autonomous social relationships. However, face to this new paradigm, users remain suspicious. They fear the violation of their privacy and revelation of their personal information. Without reliable mechanisms to enhance trustworthy communications between nodes, SIoT will not reach sufficient popularity to be considered as a leading technology. Hence, trust management becomes a major challenge to ensure qualified services and guaranteed security. Several works in the literature have tried to diagnose this problem. They proposed various trust evaluation models based on different features and aggregation methods, aiming to classify benign nodes of the SIoT network. However, related works did not allow to detect malicious nodes and couldn’t identify their types of attacks. As a result, we suggest a new trust-evaluation model in a deep learning framework. This model permits to find out the type of trust-related attacks performed by malicious nodes, which will be isolated from the network in order to achieve a reliable environment. Based on authentic data, experimentation is able to prove our system performance.