Product Recommendation Method Based on User Repurchase Behavior

Repeat purchase is a common phenomenon in daily consumption decisions, and is crucial for improving the accuracy of personalized product recommendations. However, there is limited research on modeling and predicting user repurchase behaviors. Many issues need to be studied. It is notable that most of existing methods only optimize product, basket or time series level information for repurchase recommendation. They not only ignore the modeling of multi-level information fusion, but also ignore the interpretability requirement for repurchase recommendation results. To address the above issues, we propose a novel interpretable repurchase recommendation method based on dual attention mechanism. Our method combines attention mechanism and pointer generation network to extract multi-level user repurchase preference. We also propose an S-type user repurchase dynamic preference function based on information processing theory and integrate product popularity information for mixed recommendation of repurchase and new products, which improves the interpretability and recommendation accuracy of our method. The experimental results on two real world datasets show that our proposed method outperforms baselines in multiple performance indicators and the learned parameters have good interpretability. In addition, regression studies demonstrate the reliability of our S-type repurchase dynamic preference function, which further enhances the interpretability of our method from a theoretical perspective.