Towards a deep learning-driven service discovery framework for the Social Internet of Things: a context-aware approach

The Social Internet of Things (SIoT) is a new paradigm that enables IoT objects to establish their own social relationships without human intervention. A fundamental perspective of SIoT is to make socially capable objects, wherein objects can automatically share their services capability and exchange their experience with each other for the humans’ benefit. Service discovery is a crucial task that requires fast, scalable, dynamic mechanisms. This paper aims to investigate the feasibility of adopting state-of-the-art deep learning techniques to build a social structure among IoT objects and design an effective service discovery process. To achieve this goal, we propose a framework that includes three phases: i) collecting information about IoT objects; ii) constructing a social structure among IoT objects using; and iii) developing an end-to-end service discovery model using the language representation model BERT. We conducted extensive experiments on real-world SIoT datasets to validate our approach, and the experimental results demonstrate the feasibility and effectiveness of our framework.