A Context-Aware Hybrid Recommendation Framework for Real-Time IoT Environments

As things in IoT keep evolving, people's choices for what to use are influenced by their situation and can vary which calls for systems that can adapt to these changes. Here, a framework is described that uses suggestions from collaborative filtering, content-based filtering, and a deep neural network in partnership with ensemble learning. Realistically, it makes use of matrix factorization, a k-nearest neighbors’ approach, and a convolutional neural network (CNN) to organize its model. For example, by using CASAS and SmartThings data that is made open to the public, the model is trained and its performance tested by focusing on timing elements, recordings of user-device interaction and context details. In evaluation, precision, recall, F1-score, novelty, and context precision are used, helping to analyse a system from several angles. Matrix factorization and deep learning models cannot beat them, as they achieved a better accuracy (+8.5%), created more innovative recommendations (+12.3%) and operated faster on average (average latency was reduced by 27%). The architecture is equipped to handle privacy issues and is flexible to fit within edge computing. Charts and diagrams are included to demonstrate the design and how different components connect.