Probing Digital Footprints and Reaching for Inherent Preferences: A Cause-Disentanglement Approach to Personalized Recommendations
提出DISC方法，通过解耦表示学习和因果图建模，从消费者行为中分离内在偏好、商品显著性和从众效应，提升推荐准确性和可解释性，实验证明优于现有方法。
This study introduces DISC (Disentangling consumers’ Inherent preferences, item Salience effect, and Conformity effect), a novel personalized recommendation approach that leverages disentangled representation learning and causal graph modeling to provide interpretable and effective recommendations. By analyzing consumer behavior across various shopping stages, DISC identifies and differentiates the inherent factors that influence purchasing decisions. DISC cuts through biases to pinpoint consumers’ inherent preferences driving purchases, empowering platforms with the ability to deliver tailored recommendations that resonate deeply with users. Through extensive experiments on real-world data sets, DISC significantly outperforms existing methods, demonstrating its superiority in both in-sample prediction and generating recommendations that align with consumers’ true interests. With its robust performance and theoretical underpinnings, DISC holds promising implications for e-commerce platforms seeking to enhance recommendation accuracy, interpretability, and user engagement.