Leveraging fashion data to make shopping recommendations

Fashion recommendation problems are characterized by sparse datasets and large catalogs of styles that have short lifespans. Traditional transaction-based approaches such as collaborative filtering are not well suited to address that sparsity and have known limitations with new products or users (i.e., the cold start problem).

Supervised learning based on features can overcome some of these limitations, but supervised learning requires feature data. Rhonda Textor explains how to transform raw retail data into scalable recommendations using widely available machine learning libraries. She demonstrates how to extract several fashion relevant features for products and leverage consumer-provided registration data to train a machine learning algorithm to make personalized fashion recommendations.