A Transformer-based Semi-Supervised Learning for Occupant Feedback Evaluation

Indoor Environmental Quality (IEQ) focuses on ensuring that occupants experience comfortable and healthy indoor settings, covering factors such as temperature, humidity, air quality, acoustics, and lighting. Assessing IEQ from occupants’ perspectives is challenging due to the subjective natures of comfort and health perceptions. Existing methods for collecting direct occupant feedback, including surveys, questionnaires, observations, and interviews, frequently result in insufficient data, limiting the comprehensive and holistic understanding of occupant satisfaction in indoor environments. In response to these challenges, this paper proposes a semi-supervised learning-based framework aimed at deriving a metric of comfort and health satisfaction from sparse occupant feedback. Semi-supervised models leverage both labeled examples and the underlying structure of unlabeled data, enhancing their ability to generalize. The proposed framework leverages a Transformer-based model to extrapolate and analyze limited occupant feedback (labeled data) through leveraging multidimensional indoor environmental data (unlabeled data), offering a robust approach to assess subjective IEQ satisfaction. The formulation of the proposed framework involves three key steps: (1) collection of IEQ data, (2) collecting occupants’ direct feedback feedback, and (3) iterative model training and data annotation. Experimental evaluation conducted at the Virginia Tech Blacksburg campus yielded promising results, showing the effectiveness of the proposed approach in supporting IEQ management and occupant feedback analysis. This study contributes to the field of adaptive environmental controls, aimed at creating more tailored indoor environments that meet the specific comfort and health needs of occupants.