YiGraph

<div align="center"> <table border="0" cellspacing="0" cellpadding="0"> <tr> <td align="center" valign="middle" style="padding-right: 30px;"> <img src="figure/logo.png" alt="YiGraph Logo" width="180" /> </td> <td align="left" valign="middle"> <h2 style="margin: 0; font-size: 24px; font-weight: 600; color: #2c3e50;">End-to-End Intelligent Graph Data<br/>Analysis Agent System Based on AAG Framework</h2> </td> </tr> </table> <p style="margin-top: 20px;"> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="License"></a> <a href="https://www.python.org/downloads/"><img src="https://img.shields.io/badge/python-3.11+-blue.svg" alt="Python"></a> <a href="http://iDC-NEU.github.io/YiGraphDocs/"><img src="https://img.shields.io/badge/📚-Docs-purple.svg" alt="Docs"></a> <a href="#-contact-us"><img src="https://img.shields.io/badge/📞-Contact_Us-green.svg" alt="Contact"></a> </p>

English | 简体中文

</div>

---

📖 Project Introduction

YiGraph is an end-to-end intelligent graph data analysis agent system designed to help users quickly gain insights into key relationships from complex data.

YiGraph can automatically extract entities and relationships from various raw data sources such as logs, documents, and tables to build structured graph data. Users only need to describe business problems in natural language, and the system will automatically plan the analysis process, complete calculations, and generate clear, interpretable, and traceable analysis reports.

Internally, large language models are responsible for understanding user intent, breaking down analysis tasks, and organizing final outputs. The core technology supporting the reliability of analysis results is the AAG (Analytics-Augmented Generation) framework. AAG treats analytical computation as a core capability, invoking graph algorithms and graph systems at key stages to complete verifiable calculations, which are then interpreted and summarized by the model.

Therefore, YiGraph is not just a conversational AI that "answers questions", but an intelligent graph analysis agent that can transform business problems into executable and reviewable analysis processes.

Applicable Scenarios

YiGraph can flexibly adapt to different industries and business needs, covering various complex relational data analysis scenarios, including but not limited to:

• Financial anti-money laundering and suspicious transaction analysis: Automatically build transaction networks from massive transaction flows to identify abnormal fund paths and suspicious transaction loops
• E-commerce risk control and wool party identification: Integrate multi-source data such as accounts, devices, and addresses to build graphs and discover organized fraud and associated malicious behavior
• Enterprise association and risk investigation: Build graphs through enterprise, equity, and transaction relationships to penetrate complex structures and identify potential compliance and operational risks
• Park/city event analysis: Unify access control, trajectory, and event data into graphs to restore personnel relationships and event evolution processes
• Supply chain risk analysis: Integrate enterprise and transaction data to build supply chain networks, locate hidden associated risks and transmission paths

---

⚡ Core Features

1. Knowledge-Driven Task Planning

The system first understands what the user's question "wants to solve", then breaks it down into executable analysis steps:

• What data fields and relationships are needed
• What kind of graph should be built (which entities, which relationships)
• What analysis methods and parameters should be used
• How analysis results should be interpreted and presented

You don't need to understand graph algorithms; the system will translate "what I want to query" into "how to do the analysis".

2. Algorithm-Centric Reliable Execution

YiGraph will not let the model arbitrarily "write a piece of uncontrollable code and run it". Instead, it centers on "verifiable algorithm modules" for invocation and combination, making each analysis step:

• Reproducible: Same input yields stable and consistent output
• Traceable: Know which algorithms were used and which steps were executed
• More reliable: Key calculations are completed by professional modules rather than pure text reasoning

3. Task-Aware Graph Construction

YiGraph will not indiscriminately build all raw data into one large graph. It will selectively extract and construct "entities and relationships relevant to the problem" based on current task needs, avoiding interference from irrelevant structures, and organize the graph into a form more suitable for execution, thereby improving efficiency and result quality.

4. Rich Graph Algorithm Library

Built-in 100+ graph algorithms covering 11 major categories, providing professional algorithm support for various graph analysis scenarios:

| Algorithm Category | Number of Algorithms | Typical Algorithms | Application Scenarios | |---------|---------|---------|---------| | Basics | 10 | BFS, DFS, Topological Sort, DAG Detection, Ancestor/Descendant Query | Graph structure validation, dependency analysis, hierarchical traversal | | Path | 13 | Dijkstra, Bellman-Ford, Floyd-Warshall, Eulerian Path, DAG Longest Path | Path planning, relationship chain analysis, critical path | | Centrality | 14 | PageRank, Betweenness Centrality, Closeness Centrality, Eigenvector Centrality, HITS, VoteRank | Key node identification, influence assessment, seed selection | | Connectivity & Components | 13 | Connected Components, Strongly Connected Components, Cut Vertices/Edges, Minimum Cut, Node/Edge Connectivity | Network robustness, vulnerability analysis, island identification | | Clustering & Community | 17 | Louvain, Leiden, Label Propagation, k-clique, Girvan-Newman, Clustering Coefficient, Cycle Detection | Circle identification, gang discovery, tightness analysis | | Tree & Spanning Tree | 3 | Minimum Spanning Tree, Maximum Spanning Tree, Random Spanning Tree | Network skeleton extraction, cost optimization | | Flow & Cut | 5 | Edmonds-Karp, Maximum Flow, Minimum Cut, Gomory-Hu Tree | Capacity planning, bottleneck analysis, network resilience | | Matching & Coloring | 6 | Maximum/Minimum Weight Matching, Greedy Coloring, Minimum Edge Cover | Resource allocation, conflict detection, task scheduling | | Cliques & Cores | 4 | Maximal Clique Enumeration, Maximum Weight Clique, k-core, Core Number | Tight group discovery, core member identification | | Distance & Measures | 8 | Eccentricity, Diameter, Radius, Center/Periphery, Wiener Index, Assortativity Coefficient | Network health check, topology comparison, structural preference analysis | | Graph Query | 8 | Node Query, Relationship Filtering, Neighbor Query, Path Query, Common Neighbors, Subgraph Extraction, Aggregation Statistics | Data retrieval and filtering, interactive exploration, risk control investigation |

For detailed algorithm descriptions and usage guides, please refer to 📚 Online Documentation

5. Flexible Data Support

Supports multiple data source inputs:

• Graph Data
• Text Data: Documents, logs, reports, and other unstructured data

The system will automatically extract entities and relationships from raw data to build structured graph data.

6. Multiple Operating Modes

• Normal Mode: Users only need to submit their business questions. YiGraph will automatically parse the problem, select appropriate graph algorithms, execute the computation, and generate an analysis report. This mode is suitable for non-technical or general business users.
• Interactive Mode: Users collaborate with YiGraph to analyze business problems. For a given question, YiGraph interacts with the LLM to determine the computation workflow and graph algorithms, then executes the plan and returns an analysis report. This mode is suitable for advanced users who are familiar with both the business and graph algorithms.
• Expert Mode: Users directly specify the business problem along with the solution approach, computation steps, and graph algorithms. YiGraph then executes the provided plan and returns an analysis report. This mode is intended for expert users with deep knowledge of the business and graph algorithms.

---

🎯 Version Release

v0.1.0 (Current Version)

Core Capabilities

• ✅ Complete graph computing engine (based on NetworkX and Neo4j)
• ✅ Intelligent task planning and execution
• ✅ 100+ graph algorithms support, covering 11 major categories
• ✅ Multi-data source support (graph/text)
• ✅ Interactive dialogue interface

Roadmap

v0.2.0 (Planned)

• 🔄 Expand the graph algorithm library to 200–300 algorithms
• 🔄 Add an integrated graph learning module (training/inference)

---

🚀 Quick Start

1. Environment Preparation

1.1 Python Version Requirements

• Python >= 3.11

Please confirm that the current Python version meets the requirements:

python --version
# or
python3 --version

1.2 Create Virtual Environment with Conda (Recommended)

conda create -n AAG python=3.11
conda activate AAG

1.3 Neo4j I