Ariadne: A Multi-Agent Collaborative System for Interactive Literature Analysis and Research Support

ACL ARR 2025 May Submission5443 Authors

Abstract: The rapid expansion of scholarly publications has resulted in severe information overload, posing significant challenges for researchers in retrieving, evaluating, and synthesizing scientific knowledge.While large language models (LLMs) have shown potential in assisting scientific workflows, existing approaches often suffer from hallucinations and lack support for iterative, exploratory research.We introduce **Ariadne**, a multi-agent collaborative system designed for interactive literature analysis. Ariadne dynamically adapts to evolving research intents in the course of user interaction, employs flexible retrieval strategies, and performs hierarchical evidence synthesis to more effectively address complex scientific queries.Experiments on single-turn scientific QA benchmarks, including **SciFact** and **SCHOLARQA-MULTI**, demonstrate state-of-the-art performance. Moreover, human evaluations in real-world research scenarios indicate that **Ariadne** delivers superior performance compared to existing baselines.

Paper Type: Long

Research Area: Dialogue and Interactive Systems

Research Area Keywords: retrieval; knowledge augmented; multi-modal dialogue systems;

Contribution Types: NLP engineering experiment, Reproduction study, Publicly available software and/or pre-trained models

Languages Studied: English; Chinese

Submission Number: 5443