2024-06-11 –, Kesselhaus
Everyone is unique, and that is especially true for what they eat. At Albert Heijn, we are moving away from popularity-based search for the broader customer base, in favor of tailoring search to our unique customers' tastes. We will tell you all about this transition.
Albert Heijn, the largest supermarket chain of the Netherlands, has a loyalty program, called the Bonus Card, allowing us to tie all purchased products to a customer whenever they scan the card, either in the store or online. This creates a huge potential for personalisation, which had previously not been utilised within product search. We will present about our journey going from popularity-based search for the broader customer base, to a tailored search experience for all of our unique customers using the information we gather from the Bonus Card. Specifically, we will focus on Learning-to-Rank (LTR). This transition was definitely not without it's challenges, on which we would love to share our experience:
* Handling large quantities of data. Going from aggregated popularity to single user relevancy meant a million-fold increase in the quantities of data that we were handling.
* Handling large amounts of point-in-time accurate features in offline feature stores.
* Using distributed computing to train a model on this large quantity of data.
* Redefining the concept of relevancy. How can we incorporate profitability?
* Handling position bias in our data.
* Using Kafka to facilitate the quick transfer of offline features to online features during inference.
This presentation is relevant for anyone who is struggling to go from legacy popularity-based search to personalised search for big customer bases.
You will learn how to face the challenges of moving to large quantities of personalized data, distributing a model to learn on this exploded quantity of data, and redefining the concept of relevancy.
My name is Luuk Kaandorp, 25 years old, and I've been working at Albert Heijn as a Data Scientist in the Search time for about 1.5 years. I previously did a Bachelor in Information Sciences at the Vrije Universiteit Amsterdam (VU), and a Master in Artificial Intelligence at the Universiteit van Amsterdam (UvA). During my master's, I specialised mainly in Natural Language Processing and Information Retrieval. After my master's thesis on Diversity in Personlised News Recommendations at RTL, I joined Albert Heijn as my first full-time job. Since then, I've been trying to apply all the knowledge I gained during my studies on the product search domain, to deliver our customers with the most relevant search experience on our website and app.
Vincent Peijnenburg is a Data Scientist in the Search team at Albert Heijn, the largest supermarket chain in the Netherlands, both offline and online. He has been in this position for about 1 year and has previously worked for Transavia (the Dutch airline part of KLM-group) for 5 years, working on recommender systems, pricing systems, and other ML applications in the commerce domain. He has experience with both the modeling and the Mlops side, delivering end-to-end solutions, and is always eager to try out the latest tools and techniques and apply them in a business context.