Summary

YiGraph is an end-to-end intelligent agent system simplifying complex graph data analysis. It empowers users to derive insights from diverse data sources by automating entity extraction, relationship building, analysis planning, and report generation via natural language. The system transforms business problems into executable, reviewable analysis processes, enhancing reliability and interpretability.

How It Works

YiGraph utilizes the Analytics-Augmented Generation (AAG) framework, integrating analytical computation as a core capability. LLMs interpret user intent and plan analysis steps, while verifiable graph algorithms execute key calculations. This ensures reproducibility and traceability, moving beyond pure text reasoning for reliable results. Task-aware graph construction selectively builds relevant data structures, improving efficiency and output quality.

Quick Start & Requirements

• Installation: Clone repository (git clone https://github.com/iDC-NEU/YiGraph.git), cd YiGraph, pip install -r requirements.txt.
• Environment: Python >= 3.11 (Conda 3.11 recommended).
• Database: Neo4j 3.5.25 required (must be running); requires Java 8 or 11.
• Configuration: Edit config/engine_config.yaml (LLM, API keys) and config/data_upload_config.yaml (dataset paths).
• Execution:
  • Web Interactive: python web/frontend/run.py (Recommended)
  • Terminal Interactive: python aag/main.py
• Documentation: http://iDC-NEU.github.io/YiGraphDocs/

Highlighted Details

• Key Features: Knowledge-driven task planning, algorithm-centric reliable execution, task-aware graph construction.
• Algorithm Library: 100+ graph algorithms across 11 categories (Path, Centrality, Connectivity, Clustering, etc.).
• Data Support: Graph data, text data (documents, logs).
• Operating Modes: Normal (automated), Interactive (user-guided), Expert (user-defined plan).
• Current Version (v0.1.0): Graph engine (NetworkX, Neo4j), intelligent planning, multi-data support, dialogue interface.
• Roadmap (v0.2.0): Planned expansion to 200-300 algorithms and graph learning module.

Maintenance & Community

Authored by Qiange Wang et al. Contributions welcomed. Community channels include WeChat, Xiaohongshu, Twitter. Roadmap for v0.2.0 outlined.

Licensing & Compatibility

MIT License: Permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

Requires specific Python (>=3.11), Java (8/11), and Neo4j (3.5.25) versions. GPU likely needed for embedding models. LLM API keys are necessary. v0.1.0 suggests an early-stage release.