At H&M Group, we are on an exciting journey where fashion and tech make magic together. As new rapid customer behaviors are transforming the fashion retail industry, our mission is to meet and exceed our customers’ expectations. The solution we propose to address this challenge is to combine multiple data sources such as click streams, images, and textual descriptions in unified numerical representations. These numerical representations, also called vectorial embeddings, are created by using encoders of different data sources with the goal to translate product attributions into a vectorized shape. In this talk, we will present our strategy for obtaining embeddings and outline in detail some of the learnings and challenges we have had.
Our strategies of encoding fashion into vectors.
How we are combining data sources to improve our embeddings.
The process of choosing data and model depending on the challenge.