GCIB: Graph Contrastive Information Bottleneck for Multi-Behavior Recommendation

With the rapid emergence of multi-behavior learning in recommender systems, leveraging auxiliary user behaviors has proven effective for mitigating target-behavior data sparsity. Yet auxiliary behavior graphs frequently contain noisy or irrelevant interactions that do not align with the target task, impeding the learning of accurate user and item embeddings. Moreover, the scarcity of direct supervised from the target behavior complicates the extraction of informative collaborative signals. In this paper, we introduce GCIB Graph Contrastive Information Bottleneck, a novel framework that denoises auxiliary behavior information and enriches target behavior representations at both the structural and feature levels. At the structural level, GCIB employs a Graph Information Bottleneck (GIB) objective to maximize mutual information between the denoised auxiliary graph and the target-behavior graph while minimizing mutual information with the original auxiliary graph. This formulation preserves task-relevant structural patterns and suppresses spurious interactions. At the feature level, we propose a cross-behavior Graph Contrastive Learning (GCL) scheme in which denoised auxiliary features and target-behavior features serve as complementary views for both users and items. By contrasting these views, GCIB enriches sparse target-behavior representations with semantics distilled from auxiliary behaviors. Extensive experiments on multiple real-world datasets demonstrate that GCIB outperforms state-of-the-art baselines, highlighting its ability to learn noise-resilient and target-aware representations for multi-behavior recommendation.

When using online platforms, people might click on or save many items, but only purchase a few. Because actual purchases are rare, recommendation systems use these extra actions to predict what users truly want. However, this supplementary data is often messy: casual clicks or accidental views can confuse the system with irrelevant "noise." To solve this, we developed GCIB, a framework that acts as a smart information filter. It sifts through a user's various online actions to separate meaningful clues from random noise. First, it analyzes the underlying patterns connecting casual actions to actual purchases, actively discarding the unhelpful data. Then, it contrasts these refined clues with the user's buying history to build a much clearer profile of their true preferences. By effectively filtering out noise and amplifying useful signals, our approach delivers significantly more accurate recommendations. This ensures that the products suggested to users are the ones they genuinely want, improving the overall online experience.