Prototype-Based Explanations for Interpretable Fashion Recommendations

As recommender systems become integral to online retail platforms, understanding why an item is recommended is increasingly important. This is particularly challenging in fashion, where user preferences are subjective and context-dependent. This thesis explores prototype-based explanations as a means of enhancing interpretability in fashion recommendation systems. Using the large-scale H&M Personalized Fashion Recommendations dataset, we aim to develop hybrid models that combine collaborative signals with rich item metadata. To generate explanations, we aim to introduce a method that associates each recommendation with a small set of prototypical items the user has interacted with and highlighting similarities in attributes such as color, garment group, or style.