Optimization of Personalized Fashion Recommendations for H&M: A Collaborative Filtering Algorithm Approach with Temporal Time Interval Analysis

Abstract

This study presents a personalized fashion recommendation system for the H&M dataset, utilizing a cosine similarity-based collaborative filtering algorithm. This study investigates the effect of temporal segmentation on recommendation performance by conducting three experiments using datasets divided into two-week, one-month, and two-year time intervals. The experimental results show that the two-year interval achieves the best performance, producing a Mean Average Precision (MAP) of 0.02254 with a computational time of 2741.7 seconds. In contrast, the two-week interval achieves a MAP of 0.00915 in 1609.2 seconds, while the one-month interval produces a MAP of 0.00554 with a computational time of 3118.9 seconds. The main contribution of this study lies in the optimization of data structure transformation through dictionary-based modeling, which significantly improves training efficiency. These findings underscore the crucial role of temporal granularity in improving the accuracy and computational efficiency of collaborative filtering-based personalized fashion recommendation systems.