A Behavior‐Aware Group Food Recommendation Framework Based on Novel Adaptive Aggregation and Deep Preference Learning
Reetu Singh, Pragya Dwivedi, Vibhor Kant · Concurrency and Computation Practice and Experience · 2026
ABSTRACT Food recommender systems (FRSs) help users to discover food content, products, or services based on their preferences. Most FRSs are designed for individuals, but in many real‐world situations, decisions are made collectively, such as friends planning a party, families choosing a restaurant, or roommates buying groceries. In these cases, the system needs to combine multiple users' preferences into a single group food recommendation. Traditional aggregation methods struggle when there are very few ratings or when some ratings are extremely high or low. They treat outliers and extreme ratings the same as normal ones, which can make the group score wrong. To address this limitation, our proposed adaptive aggregation methods automatically reduce the impact of unreliable ratings, leading to more stable group preferences. In this paper, a novel group food recommender framework is proposed in three distinct phases. It starts with the group formation phase, where groups are generated based on users' rating similarity and rating behavior, with a random grouping method serving as a baseline. In the second phase, user preferences are merged using three novel aggregation techniques: Adaptive Trimmed Mean, Hybrid Trim, and Adaptive Percentile to limit the influence of extreme ratings. Finally, our proposed framework utilizes deep matrix factorization to generate group food recommendations, which efficiently captures nonlinear user‐item interaction within group preferences. The performance of our proposed framework is evaluated using RMSE for rating prediction and F1‐measure for top‐N recommendation on two benchmark datasets, namely Allrecipes.com and Food.com. The proposed adaptive aggregation with DeepMF consistently outperforms standard baselines, demonstrating effective support for group decision‐making in food recommendation.