講者:陳釗而教授 (Department of Economics, Senshu University)
演講主題:Leveraging Predictive Models for Causal Targeting on Digital Platforms
講題摘要:
In collaboration with a leading Taiwanese e-commerce entity, we conducted A/B tests to assess the impact of re-engagement prompts on abandoned cart recoveries, focusing on the development of personalized targeting policies from a causal perspective. Unlike conventional approaches that rely on predictive machine learning methods, we utilize causal machine learning techniques, particularly the Causal Forest, to estimate the conditional average treatment effect and develop targeted interventions. Additionally, we address the challenges posed by the lack of specific user demographic data on most e-commerce platforms. By leveraging behavioral data and employing multilayer perceptron (MLP) deep learning models and the Hidden Markov model, we construct estimated covariates and latent states to refine targeting strategies. Our findings indicate that the hybrid approach, combining the strengths of predictive and causal machine learning, results in the most effective targeting. This integrated procedure not only enhances the precision of targeting but also contributes to higher overall welfare, signifying the potential of combining predictive and causal models in digital marketing strategies.