Dual Conditional Diffusion Models and Generative Diffusion Models for Sequential Recommendations

Abstract

Sequential recommendation systems are designed to forecast the subsequent item a user is expected to engage with, based on their past interactions. Both Generative Diffusion Models for Sequential Recommendations (known as DiffuRecSys) and Dual Conditional Diffusion Models for Sequential Recommendation (referred to as DCRec) utilize diffusion models to enhance the accuracy of recommendations. While DiffuRecSys emphasizes improving robustness and understanding user-item interactions via cross-attention and offset noise, DCRec adopts a dual conditional strategy that combines both implicit and explicit conditioning to boost recommendation accuracy and computational efficiency. This paper presents a comparative evaluation of the two methods, emphasizing their approaches, significant contributions, and findings. Both models show remarkable advances compared to leading baseline methods, with DiffuRecSys particularly adept at understanding varied user preferences, while DCRec stands out in terms of both accuracy and efficiency. The overview wraps up with an examination of their individual advantages, drawbacks, and possible paths for future development.