Binyam Gebre

Representation Learning in e-Commerce Recommender Systems

How do you represent millions of e-commerce customers with their multiple interests? How do you represent millions of products such that customer interests can efficiently be matched to products? I will address these questions in my talk. I will highlight that 1) energy-based models, which are models for handling multiple predictions, are well suited to learning customer and product representations and that 2) with approximate nearest neighbor search, we can do efficient inference at scale. At, the largest e-commerce company in the Netherlands, we implemented this approach to personalize promotional products and obtained significant conversion uplifts.

Binyam Gebre has a background in computer science and artificial intelligence. Since he completed his PhD in 2015, he has been working on diverse industry applications involving machine learning, natural language processing and computer vision systems. Currently, he is working as a senior data scientist at, building large-scale recommender systems using deep learning (e.g. personalized product feeds). Prior to his current role at, he has worked as a deep learning scientist for Philips Research in the areas of medical image understanding and personal health, contributing to a product with embedded deep learning (perfectcare-9000-series-steam-generator-iron).

This website uses cookies to ensure you get the best experience. Learn more