Reproduce “Benchmarking for Deep Uplift Modeling in Online Marketing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dugang Liu, Xing Tang, Yang Qiao, Miao Liu, Zexu Sun, Xiuqiang He, Zhong Ming
Researchers introduced DUMOM, an open benchmarking framework for Deep Uplift Modeling (DUM) in online marketing, to standardize evaluation across 13 models on Criteo and Lazada datasets. Their findings revealed that no single DUM model consistently outperforms others, and model performance is highly sensitive to data preprocessing choices and test distribution characteristics.
Key insights
- There is no single, consistently superior Deep Uplift Modeling (DUM) model across all datasets and preprocessing configurations, indicating performance is highly context-dependent.
- DUM models exhibit significant sensitivity to data preprocessing choices, specifically instance deduplication and feature normalization, which can either improve or degrade performance depending on the dataset and model.
- The generalization capabilities of current DUM models are limited when faced with varying test distributions (e.g., selection-biased vs. unbiased Randomized Controlled Trial environments).
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.