End-to-End Optimization of Quantization-Based Structure Learning and Interventional Next-Item Recommendation

Kairui Fu, Qiaowei Miao, Shengyu Zhang, Kun Kuang, Fei Wu

Abstract: With the development of deep learning, more and more related techniques are used in recommender system, making it more effective and reliable. However, due to the various distribution of real-time data, those deep-learning-based methods can merely learn the correlation between data rather than the actual causal effect, decreasing the performance of recommenders when a distribution shift occurs. Therefore, causal structure learning, which has been proposed to search for causal relationships between variables, is applied in recommender systems. However, existing methods assume that the recommender system is a non-interventional environment, making the causal graph learned not entirely correct. In this paper, we propose to decouple the recommender module and the causal module to consider the intervention of recommender system when building a causal graph. We utilize vector quantization to learn a cluster-level graph rather than an item-level graph to guarantee an acceptable training time. With an adjustable number of clusters, our model can adapt datasets of any size and be trained within a certain period. We conduct extensive experiments on both real-world and synthetic OOD datasets to demonstrate that our model is more effective than other state-of-the-art sequential recommenders.