Intelligent Device Disambiguation for Smart Home Control

Abstract

User interaction with smart devices is challenging when there are multiple devices that can perform the required task. Disambiguating between similar devices is a common problem that user faces when controlling devices from an app or from voice enabled smart assistants, because of the time required to interact. Moreover user command might be incomplete in case of smart assistants, leading to further challenges in identifying the user intended device. To predict the user intended device, we propose a machine learning based device disambiguation service using XGBoost algorithm. The predictions are based out of historical usage pattern of smart home users and is personalized for them. The machine learning model is optimized using random search over hyper-parameters in a completely automated fashion, which ensures optimum user experience. The solution addresses the important problem of identifying the device intended by the user and is a suitable platform for further improvements in voice assistant enabled smart home experience.