MHPAG: A Multi-Modal Hierarchical Pairwise Attention Graph for Personalized Clothing Recommendation

In this era, the rapid growth of e-commerce led for the demand of intelligent and personalized clothing recommendation systems to improve customer satisfaction and minimize decision fatigue. Traditional methods that mainly depend on collaborative filtering and content-based methods are confronted with issues like cold-start, limited context awareness, and poor scalability. This paper proposes a novel Multi-Modal Hierarchical Pairwise Attention Graph (MHPAG) framework to address these shortfalls. The proposed model utilizes probabilistic user attribute embeddings, temporal interaction encoders, multimodal feature fusion, and relational graph attention networks to capture both user preference and the relationships of the items at a higher scale. To address the cold-start and sparsity issues noted, counterfactual augmentation, and meta-learning are employed. The proposed model substantially outperformed existing methods with an accuracy of 93.32 %, precision of 94.21 %, recall of 93.54 %, and F1 score of $\mathbf{9 3. 2 1} \boldsymbol{\%}$. Thus, MHPAG demonstrates its suitability to provide a capable, personalized, and context-aware fashion recommendation system for the development of fashion recommendation systems.