Mingtao Xu, Wei Wei, Jun Zhao, Zhicong Cheng, Huakang Niu · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3841005
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).
Jointly modeling search and recommendation has emerged as a promising paradigm to capture diverse user preferences. However, the undifferentiated integration of cross-domain behaviors often induces negative transfer, driven by intrinsic intent ambiguity and extrinsic cross-domain semantic interference. To address this, we propose DiffuSAR, a diffusion-based framework designed to purify cross-domain behavioral sequences. First, we introduce a dual-branch intent disentanglement module to regularize domain-specific behaviors into discrete intent prototypes, which serve as explicit semantic anchors for the denoising process. Second, we devise a masked autoregressive denoising mechanism to efficiently suppress cross-domain semantic interference, leveraging preceding target-domain behaviors as guidance signals. Finally, we incorporate a trajectory loss to guide the reverse diffusion trajectory, progressively aligning cross-domain representations with target intents in a coarse-to-fine manner. Extensive experiments on two public datasets demonstrate that our approach consistently outperforms state-of-the-art baselines on both recommendation and search tasks, validating the efficacy of generative denoising in mitigating negative 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
No comments yet — start the discussion below.