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

TL;DR

ScienceDB AI uses LLM to recommend scientific datasets, achieving 30% accuracy improvement.

cs.IR 🔴 Advanced 2026-01-03 2 views
Qingqing Long Haotian Chen Chenyang Zhao Xiaolei Du Xuezhi Wang Pengyao Wang Chengzan Li Yuanchun Zhou Hengshu Zhu
dataset recommendation scientific data conversational recommendation agent recommender data sharing

Key Findings

Methodology

ScienceDB AI employs a Scientific Intention Perceptor, Structured Memory Compressor, and Trustworthy Retrieval-Augmented Generation framework for dataset recommendation. The Perceptor extracts experimental elements from complex queries, the Compressor manages multi-turn dialogues, and the Trustworthy RAG provides citable dataset references through a two-stage retrieval mechanism.

Key Results

  • ScienceDB AI achieved over 30% improvement in offline metrics compared to advanced baselines and a 200% increase in click-through rates in online A/B tests.
  • Through experiments with over 10 million real-world datasets, the system significantly enhanced accuracy and trustworthiness in dataset recommendations.
  • Eliminated the shortcomings of traditional systems in understanding complex scientific intents.

Significance

ScienceDB AI addresses the shortcomings of traditional recommendation systems in handling complex scientific dataset needs, promoting sharing and utilization of scientific data. Its innovative conversational recommendation system understands researchers' scientific intents, providing personalized recommendations and enhancing dataset usability and reproducibility.

Technical Contribution

ScienceDB AI introduces a Scientific Intention Perceptor and Structured Memory Compressor, enhancing the understanding of complex queries. Its two-stage retrieval mechanism and use of CSTR identifiers ensure traceability and trustworthiness of recommended datasets.

Novelty

ScienceDB AI is the first LLM-driven conversational recommender system specifically designed for large-scale scientific dataset sharing services, overcoming traditional systems' limitations in handling complex scientific intents and multi-turn dialogues.

Limitations

  • The system may encounter errors when handling extremely complex scientific intents, especially in interdisciplinary fields.
  • The dialogue system's limited context window may lead to information forgetting.

Future Work

Future research could further optimize the dialogue system's memory capabilities and expand to more disciplines to enhance the system's universality and accuracy.

AI Executive Summary

With the rapid development of AI for Science, the importance of high-quality scientific datasets in accelerating discoveries has become increasingly prominent. However, existing dataset recommendation systems fall short in handling complex scientific intents and multi-turn dialogues. ScienceDB AI addresses these issues by introducing a Scientific Intention Perceptor, Structured Memory Compressor, and Trustworthy Retrieval-Augmented Generation framework. Its innovative conversational recommendation system understands researchers' scientific intents, providing personalized recommendations and significantly enhancing dataset usability and reproducibility. In experiments, ScienceDB AI achieved over 30% improvement in offline metrics compared to advanced baselines and a 200% increase in click-through rates in online A/B tests. Nevertheless, challenges remain in handling extremely complex scientific intents. Future research directions include further optimizing the dialogue system's memory capabilities and expanding to more disciplines.

Deep Analysis

Background

Sharing and utilizing scientific datasets is crucial for advancing scientific discovery. In recent years, governments and research institutions have established national scientific data centers and sharing platforms. However, traditional recommendation systems fall short in handling complex scientific intents and multi-turn dialogues.

Core Problem

Existing dataset recommendation systems cannot effectively handle the complex needs of scientific datasets, particularly in understanding researchers' scientific intents and multi-turn dialogues.

Innovation

ScienceDB AI achieves understanding of complex scientific intents and personalized recommendations through a Scientific Intention Perceptor, Structured Memory Compressor, and Trustworthy Retrieval-Augmented Generation framework.

Methodology

  • �� Scientific Intention Perceptor extracts experimental elements from complex queries.
  • �� Structured Memory Compressor manages multi-turn dialogues to prevent information forgetting.
  • �� Trustworthy RAG framework provides citable dataset references through a two-stage retrieval mechanism.

Experiments

Experiments with over 10 million real-world datasets compared ScienceDB AI's performance with traditional recommendation systems, showing significant improvements in accuracy and click-through rates.

Results

ScienceDB AI achieved over 30% improvement in offline metrics compared to advanced baselines and a 200% increase in click-through rates in online A/B tests.

Applications

ScienceDB AI can be used for dataset recommendations in scientific research, helping researchers quickly find datasets that meet their needs.

Limitations & Outlook

The system may encounter errors when handling extremely complex scientific intents, especially in interdisciplinary fields.

Plain Language Accessible to non-experts

Imagine you're in a library looking for books. Traditional methods are like searching the shelves for keywords, while ScienceDB AI is like a smart librarian who understands your research intent and recommends the best books. It not only remembers your previous requests but also adjusts recommendations based on your changing needs.

ELI14 Explained like you're 14

Imagine you're playing a super complex game and need to find specific items to advance. ScienceDB AI is like a super helpful assistant who understands what items you need and helps you find them in the game. It remembers your past choices and adjusts suggestions based on your game strategy.

Glossary

Scientific Intention Perceptor

A component that extracts experimental elements from complex queries.

Used to understand researchers' scientific intents.

Structured Memory Compressor

A component that manages multi-turn dialogues to prevent information forgetting.

Maintains context in conversational recommendations.

Trustworthy RAG Framework

Provides citable dataset references through a two-stage retrieval mechanism.

Ensures traceability and trustworthiness of recommended datasets.

CSTR Identifier

A system that provides unique identifiers for datasets.

Ensures datasets are citable.

Multi-turn Dialogue

The process of exchanging information multiple times in a conversation.

Used to understand complex scientific intents.

Open Questions Unanswered questions from this research

  • 1 How to improve the system's understanding of complex scientific intents in interdisciplinary fields.
  • 2 How to further optimize the dialogue system's memory capabilities to prevent information forgetting.

Applications

Immediate Applications

Scientific Research

Helps researchers quickly find datasets that meet their needs, improving research efficiency.

Data Sharing Platforms

Enhances the dataset recommendation capabilities of platforms, promoting the sharing and utilization of scientific data.

Long-term Vision

Cross-disciplinary Applications

Expand to more disciplines to enhance the system's universality and accuracy.

Abstract

The rapid growth of AI for Science (AI4S) has underscored the significance of scientific datasets, leading to the establishment of numerous national scientific data centers and sharing platforms. Despite this progress, efficiently promoting dataset sharing and utilization for scientific research remains challenging. Scientific datasets contain intricate domain-specific knowledge and contexts, rendering traditional collaborative filtering-based recommenders inadequate. Recent advances in Large Language Models (LLMs) offer unprecedented opportunities to build conversational agents capable of deep semantic understanding and personalized recommendations. In response, we present ScienceDB AI, a novel LLM-driven agentic recommender system developed on Science Data Bank (ScienceDB), one of the largest global scientific data-sharing platforms. ScienceDB AI leverages natural language conversations and deep reasoning to accurately recommend datasets aligned with researchers' scientific intents and evolving requirements. The system introduces several innovations: a Scientific Intention Perceptor to extract structured experimental elements from complicated queries, a Structured Memory Compressor to manage multi-turn dialogues effectively, and a Trustworthy Retrieval-Augmented Generation (Trustworthy RAG) framework. The Trustworthy RAG employs a two-stage retrieval mechanism and provides citable dataset references via Citable Scientific Task Record (CSTR) identifiers, enhancing recommendation trustworthiness and reproducibility. Through extensive offline and online experiments using over 10 million real-world datasets, ScienceDB AI has demonstrated significant effectiveness. To our knowledge, ScienceDB AI is the first LLM-driven conversational recommender tailored explicitly for large-scale scientific dataset sharing services. The platform is publicly accessible at: https://ai.scidb.cn/en.

cs.IR cs.AI cs.DL