IoT devices, as a result of technological advancements, may have different ways of operating and communicating despite having the same features. Therefore, finding a specific device among the whole of deployed devices can be a difficult task. To help find devices in an efficient and timely way, we propose a recommender system using deep learning for matching W3C Web of Things artifacts (called as WoT devices) with natural language queries. The proposal uses the Transformer algorithm to study the usage of deep learning to facilitate searching for devices, assuming that the model can be used as a recommendation tool to match WoT devices in Cyber-Physical Systems.