Keld Lundgaard is a Lead Data Scientist, working on recommender systems at Salesforce Commerce Cloud Einstein team. Prior to Salesforce, Keld was a postdoctoral fellow at Stanford University, where he developed machine learning models to improve the accuracy of surface science simulations used for screening new material compounds for batteries, fuel cells, and artificial photosynthesis. Keld holds a Ph.D. from the Technical University of Denmark.
Complete the look recommenders have been shown to both enhance the shopper experience and increase revenue for merchants. Traditionally, “looks” are curated manually; however, it is expensive and almost impossible to maintain high-quality coverage with shifting shopping trends, fast-changing catalogs, and products going out of stock.
In this talk, the speaker will show how his team has built a complete look recommender that works without curation. We combine recent developments in NLP, vision, deep learning, and active learning, to create an ontology for apparel with what clothing can go together. This allows us to constrain a collaborative filtering algorithm and create outfits that make sense. I will share our challenges and learnings along the way and hope to inspire more work on recommenders that combines collaborative filtering with knowledge graph constraints.