Recommending fashion articles
Title: Recommending fashion articles
SNIC Project: Berzelius-2022-88
Project Type: LiU Berzelius
Principal Investigator: Filip Cornell <fcornell@kth.se>
Affiliation: Kungliga Tekniska högskolan
Duration: 2022-04-13 – 2022-06-01
Classification: 10201
Homepage: https://wasp-sweden.org/
Keywords:

Abstract

In this project, we investigate the possibility of using deep multi-modal transformers to improve fashion item recommendations. Our transformer model combines different information sources, such as product images and user geographical locations. It also builds deep representations that capture aspects of collaborative filtering, as well as seasonal and evolving patterns of users and items. Initial tests have shown promising results, but more computational resources are necessary to scale our experiments and more efficiently evaluate different architectural choices and hyperparameter combinations on a large industry-provided dataset.