.png){kind=icon}
Return
Target Behavior Supervised for Cascading Contrastive Multi-Behavior Recommendation
.png){kind=icon}
PRE
AI
DOI：10.1145/3844611.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Cascading multi-behavior modeling has become an important research method in the field of multi-behavior recommendation systems. These models leverage the connections between different behaviors to infer user preferences. However, they still face several challenges: (1) Initial behavior modeling often learns noise due to lack of target behavior preference, affecting subsequent behaviors and overall recommendation performance. (2) Existing models often fail to effectively propagate target-specific preference information, rendering target behavior prediction susceptible to noise. To address the aforementioned challenges, we propose a novel framework called Target Behavior Supervised for Cascading Contrastive Multi-Behavior Recommendation (TBS-CCMR), which consists of the Target Behavior Supervision module (TBS) and the Cascading Contrastive Encoding module (CCE). Specifically, in TBS module, we utilize target behavior embeddings to supervise the aggregation process of auxiliary behavior representations, fully extracting user preference semantics from each specific behavior. Meanwhile, in CCE module, the final embeddings generated from the TBS module are used as the initial embeddings in the cascade chain. A cross-behavior feature transformation function is designed to address the loss of specific preference information in the behavior chain. Finally, the model optimizes the overall objective via a multi-task optimization approach. TBS-CCMR effectively addresses the limitations of cascading multi-behavioral recommendation models that struggle to leverage multi-behavioral data. Extensive experiments on three real-world datasets indicate that TBS-CCMR consistently outperforms various state-of-the-art recommendation methods. We release our model implementation at: https://github.com/wjf8016/TBS-CCMR.
Keywords:
Collaborative Filtering,Graph Convolutional Network,Contrastive Learning,Multi-behavior Recommendation
Journal
IF:
6.6
Papers:
1.5K
Citations:
6.2K
Organization
No organization information available
Cited Papers
No cited papers available