Recommender System: A Comprehensive Overview of Technical Challenges and Social Implications
Yiquan An, Yingxin Tan, Xi Jing Sun, Giovannipaolo Ferrari · ICCK Transactions on Sensing Communication and Control · 2024
The proliferation of Recommender Systems (RecSys), driven by their expanding deployment within Sensing-Communication-Control (SCC) architectures and explosive growth of multi-sensor data streams, has cultivated a dynamic research landscape at the intersection of artificial intelligence and cyber-physical systems. Embedded in SCC pipelines, RecSys transforms heterogeneous sensor-acquired data-spanning behavioral signals, physiological measurements, spatial context, and environmental conditions-into personalized control directives for smart homes, industrial IoT platforms, intelligent transportation systems, and precision healthcare environments. This paper comprehensively reviews RecSys foundational concepts, methodologies, and challenges from algorithmic and SCC system-integration perspectives. It categorizes RecSys solutions into five paradigms-collaborative filtering, scenario-aware, knowledge & data co-driven, large language models, and hybrid approaches-analyzing how each interfaces with sensing modalities, communication constraints, and control objectives. Five technical challenges critical to SCC deployment are then examined: robustness under sensing noise, recommendation accuracy under real-time constraints, cold-start problems in sensor-sparse environments, explainability for human-in-the-loop control, and privacy-preserving sensing and communication. The review further addresses how behavioral sensing data encodes biases that propagate into control outputs, and how human factors principles inform the design of transparent, trustworthy recommendation-driven control systems. Future directions include edge-native inference, sensor-adaptive learning, and communication-aware algorithm co-design.