Reproduce “ScienceDB AI: An LLM-Driven Agentic Recommender System for Large-Scale Scientific Data Sharing Services” 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

Qingqing Long, Haotian Chen, Chenyang Zhao, Xiaolei Du, Xuezhi Wang, Pengyao Wang, Chengzan Li, Yuanchun Zhou and 1 more

Researchers at the Chinese Academy of Sciences developed ScienceDB AI, an LLM-driven agentic recommender system for large-scale scientific data sharing, addressing the limitations of traditional methods by deeply understanding complex queries and ensuring recommendation trustworthiness. The system achieved over a 30% improvement in offline metrics and a 200% increase in click-through rate in online testing compared to existing baselines.

Key insights

  • Structured intent parsing combined with a compressed memory mechanism enables LLMs to effectively understand nuanced scientific queries and manage extended multi-turn conversations.
  • A two-stage retrieval process with CSTR integration is crucial for building trustworthy LLM-driven recommenders in scientific domains, balancing efficiency with the need for verifiable and citable recommendations.
  • Agentic LLMs can significantly outperform traditional and general LLM-based recommenders by deeply interpreting scientific intentions, moving beyond simple keyword matching to contextual, task-specific data discovery.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.