Reproduce “Separated Contrastive Learning for Matching in Cross-domain Recommendation with Curriculum Scheduling” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Heng Chang, Liang Gu, Cheng Hu, Zhinan Zhang, Hong Zhu, Yuhui Xu, Yuan Fang, Zhen Chen
The SCCDR framework from Huawei Technologies improves cross-domain recommendation matching by addressing training instability through the separation of intra- and inter-domain contrastive learning, complemented by a curriculum scheduling strategy. This approach achieved state-of-the-art performance across various datasets and demonstrated a +1.92% increase in Click-Through Rate during online A/B testing.
Key insights
- Separating intra-domain and inter-domain contrastive learning tasks is crucial for mitigating training instability caused by their inherent difficulty mismatch.
- Protecting source domain representations via a stop-gradient mechanism during inter-domain learning is vital for effective and stable knowledge transfer without degrading well-established embeddings.
- Employing a curriculum scheduler, based on negative sample difficulty, improves the robustness and calibration of representations, especially in challenging cross-domain alignment scenarios.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.