Multimodal Counterfactual Fusion with Causality-Guided Contrastive Learning for Noise-Robust Recommendation

Multimodal Recommender Systems (MRSs) leverage heterogeneous data to alleviate datasparsity, yet they remain vulnerable to spurious correlations and modality noise that obscurethe rationale behind recommendations. To address this, we propose the MultimodalCounterfactual Fusion with the Causality-Guided Contrastive Learning (MCFC) framework.MCFC integrates three core modules: a Multimodal Structural Causal Model thatmotivates the assumed causal paths of multimodal biases; a Counterfactual Feature Fusionmodule with causal effect-inspired weighting that estimates each modality’s contributionvia score perturbations and suppresses noisy features; and a Causality-Guided ContrastiveLearning mechanism that builds counterfactual in-batch negatives to enforce invariantand robust representations. Extensive experiments on four public Amazon datasets showthat MCFC consistently outperforms state-of-the-art baselines on Recall@K and NDCG@K,improving Recall@20 by 29.9% over the strongest baseline on the highly sparse Baby dataset.Ablation studies confirm the contribution of each module, evidencing MCFC’s robustnessto modality noise and strong adaptability in sparse interaction settings.