Retailrocket eCommerce Dataset is a raw, anonymized dataset from a real-world e-commerce site, containing user behavior events, item metadata, and category hierarchy for recommendation modeling.
This dataset is highly valuable for researchers and engineers focused on building and evaluating personalized recommendation systems, session-based modeling, and user behavior forecasting. Its comprehensive nature enables experimentation with advanced methods like collaborative filtering, sequence modeling, and hybrid recommendation approaches.
Use case:
Personalized Recommendation and User Behavior Analytics for E-Commerce
| Example Observation | • session_id • time_since_session_start (s) • last n events in the session (type ∈ {view, add-to-cart, transaction}) • current item_id, category_id, price • aggregates (nbr_views, nbr_cart, nbr_purchased in session) • time_of_day / weekday |
| Example Action | • Choose 1 product from a candidate set (e.g. top-K popular or retrieved by a recall model). – action space ≈ K (discrete) per step. |
| Example Reward | • 0 for normal page-view, • +1 for click on recommended item, • +5 for add-to-cart, • +item_price for purchase of recommended item (dense reward variant: discounted by steps until purchase). |