MAC: New Benchmark for Multi-Attribution CVR Prediction
A new benchmark dataset, MAC (Multi-Attribution Benchmark), has been introduced to advance conversion rate (CVR) prediction by providing labels under multiple attribution mechanisms. Previously, public CVR datasets only offered labels from a single attribution mechanism, limiting the development of multi-attribution learning (MAL) methods. The MAC dataset is the first of its kind, addressing this gap. Alongside the dataset, the authors have developed PyMAL, an open-source library that includes a wide range of baseline methods to promote reproducible research in MAL. Comprehensive experiments on MAC reveal three key insights: (1) MAL consistently improves performance across different attribution settings, especially for users with long conversion paths; (2) performance gains scale with the objective function; and (3) the dataset supports further research. The paper is available on arXiv under the identifier 2603.02184.
Key facts
- MAC is the first public CVR dataset with labels from multiple attribution mechanisms.
- PyMAL is an open-source library for multi-attribution learning baselines.
- Experiments show MAL improves performance, especially for long conversion paths.
- Performance growth scales with the objective function.
- The paper is on arXiv with ID 2603.02184.
- The announcement type is replace-cross.
- The dataset addresses a gap in public CVR datasets.
- The research promotes reproducible research in MAL.
Entities
Institutions
- arXiv