Informations
Jobs
November 01, 2022
Kaggle: Machine Learning competition for the best Multi-Objective Recommender System
What is the article about?
Major online retailers such as OTTO offer their customers millions of products to explore and purchase. However, finding the right product from such a vast selection without a little guidance can be exhausting! So we work to guide our customers to those products that best match their interests and motivations using personalised recommendations. For this reason we want to enhance our ability to forecast in real time which products each customer will want to see, add to their cart and order at any given moment of their visit.
Even though an active research community focusing on recommender systems has established itself over the last couple of years, there's still a lack of large-scale user-interaction datasets available in the e-commerce domain. As a result, newly published models risk offering insufficient scalability when applied to retailers the size of OTTO. To tackle this problem and support further research in the area of session-based recommendations, we decided to publish a large-scale dataset that we gathered from anonymised behaviour logs generated in our webshop and shopping app.
Recommender System
In our target to ensure the popularity of our dataset, we quickly realised that combining it with a fun competition might well be the best way to get thousands of research teams interested in our data! We therefore decided to launch this competition on the popular data-science platform Kaggle, providing $30,000 in prize money for the three best submissions. Our competition kicks off on 01 November 2022 and we're warmly inviting everybody who's interested in applying machine learning to real-world problems to join in the fun here. The challenge will run for three months, ending on 31 January 2023. To help you get started, we also provide a GitHub repository containing a complete dataset description and evaluation scripts.
Ground Truth
The task we're encouraging our participants to solve is to build a multi-objective recommendation model to optimise both the click-through and purchase rates of the recommended articles. Most current state-of-the-art models only optimise for CTR, so we hope this multi-objective task will serve as an exciting challenge for the ML community. Come and join in!
Want to join the challenge?
3 people like this.
1Comments
- 02.01.2023 11:07 Clock
Guten Tag,
das ist ein interessanter Wettbewerb. Wir hatten mit OTTO in den vergangenen Jahren ab und zu Kontakt bgzl. Testdatenmanagement. Wir beschäftigen uns mit Werkzeugen welche Testdaten kopieren, verfremden, Umgebungen miteinander verschmelzen und generieren, je nach Bedarf. Es gibt Überlegungen/Bestrebungen TD-Werkzeuge so zu trainieren, dass sie Produktionsumgebungen analysieren und dann synthetisch Daten aufbauen. Naturgemäß gibt es die unterschiedlichsten Meinungen dazu, ob dies zu einer befriedigenden Testdatengrundlage führen wird, oder eben nicht. Wir würden uns über eine Rückmeldung von OTTO freuen... gerne würden wir wissen ob solche Ansätze interessant/praktikabel erscheinen, oder ob unsere klassische Vorgehensweise, also Produktion entkopplen und ggf. verfremden und dann über ein TD-Bestellshop die Daten an Entwickler und Tester auszuliefern, der bessere Weg ist.
Viele Grüße nach Hamburg
Leif Diesing
Write a comment
Answer to: Reply directly to the topic
Written by
Andreas Wand
Product Manager
Philipp Normann
Senior Data Scientist
Similar Articles
Team PITAugust 13, 2024About the development of genAI assistants AskARev and Searchbuddy
04Find out (in a very special way) how the PIT team at Otto Group data.works developed the generative AI-based products AskARev and Searchbuddy.ArchitectureDevelopmentOperations
Katharina MariaJuly 24, 2024Why technical writers can rescue us
07Find out how technical writers make complex software understandable and accessible and create standards and centralized knowledge!DevelopmentOperationsWorking methods
