An NLP inspired deep recommender engine for out-of-catalog items
At Chrono24 we are dealing with the largest inventory of luxury watches worldwide, covering >10 million of watches. Many of these items are limited editions or in other ways unique. In this talk we want to recap the journey from using a collaborative filtering approach for our watch recommender system to a modern, custom deep learning approach. We show how we solved the problem to recommend unseen not-yet-in-catalog watches, by building an embedding model based on their attributes, using techniques from natural language processing.
Christian Freischlag is Lead Machine Learning Engineer at Chrono24 since 2016. Christian's focus is developing systems based on Deep Learning and operating them according to ML-Ops principles for use cases like NLP-based fraud-detection, watch (image) classification and recommender systems. Before joining Chrono24, he was a Big Data Engineer at Connexity (now Taboola) and worked in data engineering at CERN while finishing his Master's in Computer Science.