Explicit Personalized Contrastive Recommendation Based on Multibehavior Denoising

Multibehavior recommendation reflect users’ personalized preferences from multiple perspectives by exploring the dependencies among various user behaviors (such as clicks, favorites, and purchases). However, they still struggle with the sparsity of target behaviors. While some approaches have integrated contrastive learning techniques, they face two significant challenges: 1) roughly summing the contrastive tasks overlooks users’ personalized behavior patterns; and 2) unreasonable denoising methods for auxiliary behaviors disrupt the original user behavior sequence. To address these challenges, we propose a novel method called explicit personalized contrastive recommendation based on multibehavior denoising (MB-PCD). Specifically, we design an explicit personalized behavior pattern extraction module to mine users’ explicit personalized behavior patterns and behavior semantics, incorporating these insights into the contrastive task. This allows the model to more accurately capture the complex dependencies between behaviors. In addition, we create a customized denoising module tailored to the characteristics of user behaviors, effectively denoising without disrupting the original order of behaviors. Furthermore, we introduce a multibehavior information fusion module to tackle the issue of target behavior sparsity. Extensive experiments across three datasets demonstrate that our method consistently outperforms various state-of-the-art (SOTA) approaches. Analysis experiments further validate that our method effectively enhances the model’s robustness.