교차 도메인 추천 성능을 향상시키기 위해 관찰된 단일 도메인 및 교차 도메인 혼란 요인을 효과적으로 분리하고, 이러한 혼란 요인의 부정적 영향을 제거하는 동시에 긍정적 영향을 활용하는 인과적 디컨파운딩 프레임워크를 제안한다.
The core message of this article is to propose a hierarchical, generative, and causal discovery-driven approach called HJID that disentangles cross-domain user representations into domain-shared and domain-specific latent factors, ensuring joint identifiability for unique parameter recovery and improved robustness under uncertain cross-domain recommendation scenarios.
Cross-domain sequential recommendation (CDSR) aims to model user preferences by integrating and learning interaction information from multiple domains at different granularities, shifting the modeling from flat to stereoscopic.
A unified framework that adaptively enhances user representations and learns disentangled user preferences to improve cross-domain recommendation performance.