Explaining Fashion Recommendations through User Segmentation

Recommender systems often lack transparency, making it difficult for users to understand why certain items are suggested. This thesis investigates the use of user segmentation as a basis for interpretable fashion recommendations. Using the H&M Personalized Fashion Recommendations dataset, we cluster users based on their purchase behavior—such as preferred product categories, price levels, and purchase frequency—to identify distinct customer segments. Recommendations are then generated based on popular items within each segment, and explanations are provided by referencing the user’s assigned segment (e.g., “This item is popular among others who buy casual basics”).