T3 - kijouneli/EAR-MP-2025 GitHub Wiki

Title

RecSys Challenge 2025

Summary

This challenge is organized by the RecSys 2025 conference.

This challenge aims to develop new methods for predicting various user behaviors. Participants will create Universal Behavioral Profiles based on user activity like purchases, cart actions, and page visits. These profiles should help solve tasks such as churn prediction and recommendations. Participants will predict both open tasks (churn, product propensity, category propensity) and hidden tasks.

Models are mainly evaluated using AUROC. For product and category propensity, novelty and diversity are also considered. Final rankings are calculated by combining ranks from each task using the Borda count method, rewarding models that perform well across all tasks.

Note: If team formation is delayed, participation may shift to a challenge hosted by another data mining conference (e.g., CIKM 2025 or ICDM 2025).

Deliverables

Presentation(slide) and Github

Expected number of team members

More than 2 people (3 people is appropriate)

Expected duration in month

3-4 months

Data sets

  • Anonymized dataset containing real-world user interaction logs

  • Dataset Description

    • product_properties
    • product_buy
    • add_to_cart
    • remove_from_cart
    • page_visit
    • search_query

GPU cloud server

Private servers are not provided. You can use the cloud servers provided by Colab or the challenge.

Additional Information

Back to Topics Page