Gradient Norm based Deep Neural Network for Food Recommender-System
International journal of intelligent engineering and systems · 2025
Food Recommender (FR) system plays a crucial role in mitigating data overload problems by helping users identify choices based on relevance and their personalized information.However, systems face challenges when suggesting new food items or providing recommendations to new users.This research proposes a Gradient Norm based Deep Neural Network (GradN-DNN) for recommending food items that greatly align with user preferences by effectively learning and understanding feature importance.This model helps identify and focus on significant features by leveraging the magnitude of gradients during backpropagation, while GradN-DNN addresses sparsity by emphasizing impactful connections.Initially, user similarity is assessed using the Louvain method, which detects user groups and clusters them based on similarities between them for enabling precise recommendation.The user rating prediction utilizes a Bipartite Graph to effectively encode user-item interactions, enabling better use of the graph clustering system.These two stages analyze user preferences efficiently, cluster the items, and improve the quality of food recommendations and user predictions.The proposed method achieves a better accuracy of 0.08975 on the Allrecipes.comdataset compared to existing methods, such as the Heterogeneous Attention Network-based User Interests-Aware FR system (UIFRS-HA).