Personalised Diet Recommendation System Using Bandits
Arjun Subhedar, Pranav Bhutada, Ansh Avi Khanna, Ritik Kumar Gupta, Raghuram Bharadwaj Diddigi · 2024
Identifying a diet that fulfills daily nutrient requirements is crucial for maintaining good health. However, optimizing solely for nutrient requirements and health goals is not sustainable in the long run, as these recommendations may not align with users' taste preferences. Therefore, balancing nutrient requirements with users' taste preferences is paramount for long-term adherence and benefits. Additionally, promoting variety in recommended dishes is vital for the success of any dietary recommendation system. In this work, we propose a novel diet recommendation system utilizing the multi-arm bandit framework, which intelligently balances nutrient requirements and taste preferences in two phases. The first phase (offline) involves training a contextual bandit to predict nutritional food options based on user context, such as body attributes and health goals. In the second phase (online), we train a personalized bandit to refine diet recommendations based on the user's taste preferences. Through empirical analysis, we demonstrate the advantages of our proposed solution.