<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> Idea2Paper is an end-to-end research agent framework designed to systematically address the challenges in transforming raw research ideas into submission-ready academic narratives. It decomposes the research process into structured phases, focusing on the critical bottleneck of research paradigm generation. The project targets researchers and aims to provide a stable, auditable foundation for scientific discovery by generating coherent research story skeletons.

How It Works

Idea2Paper employs a modular system, with its core submodule, Idea2Story, tackling research paradigm generation. Idea2Story utilizes a pre-computation-driven framework that constructs a knowledge graph from ICLR data, enabling offline knowledge graph construction rather than runtime reasoning. This approach facilitates efficient and reliable autonomous scientific discovery by transforming underspecified ideas into structured narrative skeletons. An anchored multi-agent review system, using real review statistics as anchors, provides deterministic and auditable scores for objective feedback.

Quick Start & Requirements

• Primary install / run command: Install dependencies via pip install -r Paper-KG-Pipeline/requirements.txt. Run the pipeline using python Paper-KG-Pipeline/scripts/idea2story_pipeline.py "your research idea".
• Non-default prerequisites and dependencies: Python 3.10+. Embedding models must output 4096-dimensional vectors (e.g., Qwen/Qwen3-Embedding-8B). OpenAI-compatible Embeddings APIs are supported. Configuration requires setting SILICONFLOW_API_KEY in a .env file.
• Estimated setup time or resource footprint: Not explicitly stated, but requires API keys and potentially significant computation for knowledge graph construction and review.
• Links: Idea2Story arXiv: https://arxiv.org/abs/2601.20833, Idea2Paper ResearchGate: https://www.researchgate.net/publication/400280248_Idea2Paper.

Highlighted Details

• Knowledge Graph: Built from ICLR data, featuring Idea/Pattern/Domain/Paper nodes.
• Advanced Retrieval: Implements a three-path retrieval (Idea/Domain/Paper) with a two-stage ranking mechanism (Jaccard + Embedding).
• Anchored Multi-Agent Review: Leverages real-world review statistics for grounded, auditable, and deterministic scoring.
• Comprehensive Logging: Generates per-run structured logs for full reproducibility and auditing.

Maintenance & Community

• Community: Discord server available at https://discord.gg/FfXtbREb. WeChat communication is also supported via a QR code.
• Links: Frontend README for UI details: frontend/README.md. User Guide for advanced configuration: 📘 Need More Help?.

Licensing & Compatibility

• License type: MIT License.
• Compatibility notes: The MIT license is permissive and generally compatible with commercial use and closed-source linking.

Limitations & Caveats

The frontend UI is currently unstable and terminal execution is recommended. Certain embedding API endpoints (e.g., DashScope native) may require an adapter. Strict JSON validation in the Phase 3 Critic can be optionally skipped for performance or debugging.