Jiaming Yang, Shunmei Meng, Qianmu Li, Chentong Yu · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840845
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Cross-Domain Recommendation (CDR) effectively alleviates data sparsity by transferring knowledge from auxiliary domains. To achieve fine-grained transfer, disentangled learning is increasingly adopted to decouple user interaction motivations. However, existing paradigms predominantly rely on shallow linear assumptions that fail to decouple complex non-linear semantic noise and structural biases during knowledge transfer. Furthermore, their static training-stage debiasing exposes the inference phase to structural popularity bias and propagates confounding effects into final rankings. To address these challenges, we propose Diffusion-guided Disentanglement with Causal Debiasing for Cross-Domain Recommendation (DiDC-CDR). Specifically, DiDC-CDR extracts robust single-domain priors from global and niche subgraphs to explicitly capture long-tail interests. Conditioned on these priors, a diffusion module iteratively purifies cross-domain shared item attributes, which subsequently guide a geometric orthogonal constraint to rigorously isolate intrinsic interests from conformity. Finally, causal counterfactual intervention is executed to eliminate popularity-induced spurious correlations during inference. Extensive experiments across six diverse real-world dual-target CDR scenarios demonstrate that DiDC-CDR significantly outperforms state-of-the-art baselines. Notably, in the Elec & Cloth task characterized by a substantial domain gap, our method achieves improvements of 14.32% and 16.33% in HR@10 and NDCG@10 for the target Cloth domain, respectively, thereby providing a novel generative and causal paradigm for robust knowledge transfer.
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.