核心概念
The author proposes a novel framework, GNUM, utilizing graph neural networks and two uplift estimators to address label scarcity in individual uplift modeling.
要約
The content introduces the concept of uplift modeling and presents a novel approach using graph neural networks with two uplift estimators. It addresses the challenges of label scarcity in modeling individual uplift, showcasing superior performance over existing methods through experiments on public and industrial datasets.
Key points:
- Uplift modeling measures incremental effects of actions on users.
- Existing methods rely on individual data but struggle with hidden factors.
- Proposed GNUM framework uses graph neural networks and two uplift estimators.
- Class-transformed target and partial labels help alleviate label scarcity.
- Experimental results show significant improvements over state-of-the-art methods.
The proposed method demonstrates better performance in both regression and classification settings, showcasing the importance of incorporating social graphs in uplift estimation.
統計
Most existing methods utilize user's individual features to estimate the user uplift.
The proposed GNUM framework outperforms state-of-the-art methods under various evaluation metrics.
Experimental results show an improvement of 5% to 10% in regression setting and 12% to 25% in classification setting.
引用
"The proposed algorithms have been deployed online to serve real-world uplift estimation scenarios."