This dataset contains feature values and click feedback for millions of display ads. Its purpose is to benchmark algorithms for clickthrough rate (CTR) prediction.
Dataset construction:
The training dataset consists of a portion of Criteo’s traffic over a period
of 24 days. Each row corresponds to a display ad served by Criteo and the firstcolumn is indicates whether this ad has been clicked or not.
The positive (clicked) and negatives (non-clicked) examples have both beensubsampled (but at different rates) in order to reduce the dataset size.
There are 13 features taking integer values (mostly count features) and 26
categorical features. The values of the categorical features have been hashedonto 32 bits for anonymization purposes.
The semantic of these features is undisclosed. Some features may have missing values.
The rows are chronologically ordered.
Use case
Optimization of Bidding and Creative Selection to Maximize Net Revenue in Programmatic Advertising
| Example observations | 13 numerical variables: time since last click, conversion history, floor price, etc. 26 hashed categorical variables: user ID, device, site, campaign, creative, hour, country, etc. Derivable features: remaining budget, frequency capping, current KPIs. |
| Example actions | Discretize the bid into K levels (e.g., {€0, €0.05, €0.10, …}) (bid level, creative ID) pair → |A| ≈ K × number of creatives (typically 5–20). |
| Example reward | +1 if the impression resulted in a click (label ‘1’ in the logs), 0 otherwise Profit option: α*click – cpm_paid/1000, where α is the monetary value of a click |