What it does DeepRec is a living bibliography of deep learning papers for recommendation systems and click-through rate prediction, maintained in Chinese. It organizes PDFs into practical buckets: CTR models, matching, ranking, embeddings, multi-task learning, diversity, exploration-exploitation, and reinforcement learning. Each entry tags the conference, year, and company of origin.

The interesting bit The curation has a clear industrial slant. Roughly half the papers come from Alibaba’s labs alone, with strong representation from Google, Microsoft, Huawei, and others. The included “paper structure” diagram attempts to map how these pieces relate—a rarity in paper lists, though its clarity is untested.

Key highlights

• CTR section spans foundational (Wide & Deep, DeepFM) to 2019 architectures (FiBiNET, AutoInt)
• Match section includes the YouTube DNN paper that many systems still crib from
• Embedding section folds in graph methods (GraphSAGE, GCN) alongside product embeddings from Airbnb and Alibaba
• MTL section captures Google’s MMoE and Alibaba’s ESMM, two genuinely deployed approaches to the conversion-rate problem
• Links to related lists by 王喆 (Wang Zhe), a known recsys practitioner

Caveats

• No code, no implementations, no summaries: this is strictly a PDF collection
• README notes all materials are “organized from the internet” with a contact for takedown requests
• Update frequency and selection criteria are unclear beyond “dynamic updates”

Verdict Useful if you’re building a recsys reading group or need to trace which industrial paper proposed what architecture. Skip it if you want runnable baselines or annotated explanations—this is a filing cabinet, not a textbook.