Adaptive Multi-User Preference Aggregation Using MORL and Neural User Models for Intelligent Appliance Control

School of Computer Science and Engineering, Vellore Institute of Technology, Vellore, India

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Published online: 2 March 2026

Abstract

The emergence of intelligent models has improved user preference modelling for personalized automation. Nevertheless, energy optimization frameworks developed using multi-objective reinforcement learning (MORL) have been designed and utilized mostly for single-user problems, neglecting the increasingly multi-user nature of the real-world setting (e.g., smart homes). This proposal presents a multi-user preference aggregation system based on the MORL framework developed to aggregate, equilibrate, and incorporate divergent specifications of multiple users on appliance performance into user agreements in the smart home. Each user is represented by a model, either an artificial neural network (ANN) or a convolutional neural network (CNN), which has been constructed and trained to predict a vector output representing the user's preferences at the appliance-level, e.g., temperature, time to wash, cooling intensity. These user-related vectors of multiple users (possibly all) can then be aggregated using a central MORL model, which will learn to combine and cluster user preferences in an optimal manner, dynamically. The MORL agent will receive reward signals based on user responses to appliances; positive responses will reinforce the aggregation policy, and negative responses will incur retraining of the associated user model. The two-way feedback model represents how the preference agent learns through the environment and the user disagreement to obtain the personalized equilibrium across all dimensions of preference from multiple users (possibly all) attributes. Instead of using NLP to adapt to preferences in text- based works, the suggested method emphasizes reinforcement-driven meta-learning for the purpose of adaptive preference fusion. It allows for scalable personalization where group dynamics and conflicts can be resolved automatically. The anticipated benefits include higher collective satisfaction, faster convergence of user-specific models, and more adaptability of the system over time. By re- formulating user-centred optimization as a multi- agent coordination problem, this study expands the focus of MORL beyond energy efficiency toward a general, preference-aware intelligent control architecture.