Uplift Modelling based on Graph Neural Network Combined with Causal Knowledge

Haowen Wang, Xinyan Ye, Yangze Zhou, Zhiyi Zhang, Longhan Zhang, Jing Jiang

Keywords: uplift modeling, Graph Neural Network, Knowledge Representation, Structure Learning

TL;DR: Improve uplift modeling performance through causal knowledge representation and structural neighborhood learning

Abstract: Uplift modeling is a crucial method to estimate marketing effect modeling, which is widely used to evaluate the effect of treatment on outcomes. On the one hand, we can find the treatment with the best effect through uplift modeling. On the other hand, we can find customers who tend to make corresponding positive decisions in a given treatment. The past uplift modeling methods are mostly based on the difference-in-difference(DID) framework, combined with the machine learning model as the learner to make an estimation, ignoring the relationship and confidential information among features in uplift modeling. We propose a graph neural network-based framework combining causal knowledge as an estimator of uplift value. Firstly, we proposed a causal representation method based on conditional average treatment effect(CATE) estimation and adjacency matrix structure learning. Secondly, we proposed an uplift modeling framework based on graph convolution networks to combine the causal knowledge, which has better scalability. Our experimental results show that our method can estimate the uplift value with minor errors in the general simulation data, and its performance has also been verified in the actual industry marketing data.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Probabilistic Methods (eg, variational inference, causal inference, Gaussian processes)