Yidan Fan, Tianzi Zang, Yicong Li, Juan Li, Wenze Ma · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840834
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) mitigates sparsity by transferring knowledge from source to target domains, yet its performance heavily depends on the quality of negative samples. Uniform random sampling is dominated by easy negatives that yield vanishing gradients, while hard negative sampling based on point-wise scores may suffer from false negatives and limited coverage of the global item structure. We propose SGNCDR, a progressive framework that unifies structure-aware negative selection and diffusion-based negative generation. SGNCDR selects diverse negatives from the observed item set by accounting for the global item-space structure, and constructs user-conditioned negative representations with a diffusion model using both domain-invariant and domain-specific user signals. With difficulty-controlled guidance, SGNCDR generates negatives at multiple hardness levels and integrates them into training via an easy-to-hard curriculum to refine preference modeling while maintaining stable optimization. Experiments on benchmark datasets show that SGNCDR consistently outperforms representative CDR baselines, and analyses confirm the effectiveness of each component. The codes are available here.
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.