Attention-based Fusion for Outfit Recommendation

Abstract

Figure 1: Example outfit in the Polyvore68K dataset.Fine details, such as the heels of the sandals, the flower applique on the dress and the red pendants of the bracelet, determine that these items match nicely. These details should therefore be captured in the item representations.

Outfit Generation and Recommendation—An Experimental Study

“…Based on this model, Chen et al [2] present an industrial-scale Personalized Outfit Generation (POG) model that learns from the user-item and user-outfit interactions and generates a personalized outfit on the fly. Laenen and Moens [10] propose an attention-based fusion method for outfit recommendation which fuses the information in the product image and description to capture the most important, fine-grained product features. Other Transformer-based architectures such as BERT [4] or GPT [16] have also been used to tackle language-oriented tasks.…”

“…Even though there is a significant effort put into tackling the outfit generation and recommendation problem, to the best of our knowledge, there is no in-depth evaluation and comparison of the performance of different models on this task, including both personalized and non-personalized settings. Moreover, a lot of previous work provides results based only on open-source datasets [10], but not on real-world user data. In this paper we train and evaluate our models using datasets from Zalando 1 , one of the biggest online fashion retailers in Europe, with more than 500k articles and 32M active customers per year.…”