Skip to content

mkih76/event-driven-trading-system

BranchesTags

Repository files navigation

Event Trading Intelligence System | 事件驱动交易智能系统

AI-Powered Event-Driven Trading Analysis Platform

基于大语言模型的事件驱动交易分析系统

Python FastAPI Next.js TypeScript PyTorch License Docker

English | 中文版

English Version | 英文版

Table of Contents

  1. System Architecture
  2. Analysis Pipeline
  3. Transmission Reasoning Example
  4. Project Philosophy
  5. Core Capabilities
  6. Technical Highlights
  7. System Demonstration
  8. Tech Stack
  9. Project Structure
  10. API Reference
  11. Deployment Guide
  12. License

1. System Architecture

┌─────────────────────────────────────────────────────────────────────────────────────────────┐
│                                        USER INPUT                                            │
│                              "US-IRAN Military Conflict"                                     │
└─────────────────────────────────────────────────────────────────────────────────────────────┘
                                              │
                    ┌─────────────────────────┼─────────────────────────┐
                    │                         │                         │
                    ▼                         ▼                         ▼
┌───────────────────────────────────┐ ┌───────────────────────────────┐ ┌───────────────────┐
│  ┌─────────┐ ┌─────────┐         │ │ ┌─────────┐ ┌─────────┐       │ │ ┌─────────┐       │
│  │ Event   │ │ Trans.  │         │ │ │ Trans.  │ │ Back-   │       │ │ │ Signal  │       │
│  │ Summary │ │ Map     │         │ │ │ Details │ │ test    │       │ │ │ Report  │       │
│  └────┬────┘ └────┬────┘         │ │ └────┬────┘ └────┬────┘       │ │ └────┬────┘       │
└───────┼───────────┼───────────────┼─┼──────┼──────────┼────────────┼─┼──────┼─────────────┘
        │           │               │ │      │          │            │ │      │
        │    SSE Streaming ◄────────┘ │      │          │            │ │      │
        │           │                 │      │          │            │ │      │
        ▼           ▼                 ▼      ▼          ▼            ▼ ▼      ▼
┌────────────────────────────────────────────────────────────────────────────────────────────┐
│                                       FastAPI Gateway                                         │
│                           /api/v1/analyze  |  /api/v1/analyze/stream                         │
└─────────────────────────────────────────────────────────────────────────────────────────────┘
                                              │
                    ┌─────────────────────────┼─────────────────────────┐
                    │                         │                         │
                    ▼                         ▼                         ▼
        ┌───────────────────┐     ┌───────────────────────┐     ┌───────────────────┐
        │      STEP 1       │     │       STEP 2          │     │      STEP 3       │
        │  ┌─────────────┐  │     │  ┌─────────────────┐  │     │  ┌─────────────┐  │
        │  │ Event       │  │     │  │ Transmission    │  │     │  │ Signal      │  │
        │  │ Understanding│ │───► │  │ Analysis       │  │───► │  │ Generation  │  │
        │  │   (LLM)    │  │     │  │ (LLM + Graph)  │  │     │  │   (LLM)    │  │
        │  └─────────────┘  │     │  └────────┬────────┘  │     │  └─────────────┘  │
        └───────────────────┘     └───────────┼───────────┘     └───────────────────┘
                                              │
                    ┌─────────────────────────┼─────────────────────────┐
                    │                         │                         │
                    ▼                         ▼                         ▼
           ┌────────────────┐      ┌────────────────────┐       ┌────────────────┐
           │    Knowledge   │      │      Dynamic        │       │    Backtest   │
           │      Graph     │      │      Learning       │       │     Engine    │
           │  Industry Rel.  │      │    Feedback Loop   │       │  Confidence   │
           └────────┬───────┘      └─────────┬──────────┘       └────────┬───────┘
                    │                         │                         │
                    └─────────────────────────┼─────────────────────────┘
                                              │
                                              ▼
                                 ┌────────────────────────┐
                                 │      RAG Service       │
                                 │  Real-time Market Data  │
                                 └───────────┬────────────┘
                                             │
                                             ▼
                                 ┌────────────────────────┐
                                 │   AkShare / Tushare    │
                                 │   Financial Data API   │
                                 └────────────────────────┘

2. Analysis Pipeline

sequenceDiagram
    autonumber
    participant User as User
    participant Frontend as Frontend
    participant API as FastAPI
    participant LLM as LLM
    participant KG as Knowledge Graph
    participant BT as Backtest Engine
    participant DL as Dynamic Learner

    User->>Frontend: Submit Event
    Frontend->>API: POST /analyze/stream
    API->>LLM: Step 1: Event Understanding
    Note over LLM: Identify Type, Entities, Sentiment

    API->>LLM: Step 2: Transmission Analysis
    LLM->>KG: Query Industry Constraints
    KG-->>LLM: Return Relationship Paths

    API->>LLM: Step 3: Signal Generation
    API->>BT: Historical Similarity Match
    BT-->>API: Confidence Adjustment

    API-->>Frontend: SSE Stream Response
    Frontend-->>User: Real-time Visualization

    API->>DL: Learn New Relationships
    DL-->>KG: Update Knowledge Graph
Loading

3. Transmission Reasoning Example

3.1 Flow Diagram

graph TB
    subgraph Event["GEO-POLITICAL EVENT"]
        IRAN["IRAN blockades Strait of Hormuz"]
    end

    subgraph Supply["SUPPLY SHOCK"]
        OIL["CRUDE OIL Supply ↓15%"]
    end

    subgraph Transmission["INDUSTRY CHAIN"]
        REFIN["REFINING Cost ↑20%"]
        CHEM["PETROCHEM Cost ↑15%"]
        SHIP["MARINE Cost ↑25%"]
    end

    subgraph Demand["DEMAND IMPACT"]
        AVIATION["AVIATION Demand ↓10%"]
        TOURISM["TOURISM Demand ↓8%"]
    end

    subgraph Safe["SAFE HAVEN"]
        GOLD["GOLD ↑5-8%"]
    end

    IRAN-->OIL
    OIL-->REFIN
    OIL-->CHEM
    OIL-->SHIP
    SHIP-->AVIATION
    OIL-.->TOURISM
    OIL-.->GOLD
Loading

3.2 Transmission Chains

Path Type Rate Delay
IRAN → CRUDE OIL Supply Shock 100% 1-3d
CRUDE OIL → REFINING Cost Transmission 70% 3d
CRUDE OIL → PETROCHEM Cost Transmission 60% 7d
CRUDE OIL → MARINE Cost Transmission 80% 1d
MARINE → AVIATION Cost Transmission 70% 5d
CRUDE OIL → TOURISM Risk Aversion - -
CRUDE OIL → GOLD Safe Haven - -

4. Project Philosophy

In the globalized market, major geo-political events, policy changes, and unexpected disasters often transmit rapidly through industry chains within hours to days, triggering chain reactions in related stock prices. Traditional quantitative models relying on historical data struggle to capture these "black swan" events' atypical impacts.

This system leverages LLM's logical reasoning capability to automatically construct industry chain transmission paths from raw events, combined with knowledge graph constraints, backtest verification, and real-time market data to generate actionable trading signals.

Key Innovations

Innovation Description
LLM-Driven Reasoning Go beyond historical patterns to reason about novel events
Graph-Constrained Transmission Deterministic relationships + flexible generalization
Confidence Calibration Backtest-driven signal reliability
Self-Learning Loop Continuous knowledge graph enrichment

5. Core Capabilities

Capability Description
Event Understanding LLM auto-identifies event type (geo-political/policy/disaster/economic), core entities, market sentiment
Industry Chain Transmission Multi-level transmission reasoning (cost transmission / demand transmission / substitution effect)
Trading Signals Long/short signals with confidence levels, supporting A-shares, US stocks, ETFs
Backtest Verification Dynamic confidence adjustment based on historical similar events
Dynamic Learning Auto-discover new industry chain relationships, build feedback loop
Real-time Streaming SSE streaming display of LLM reasoning process

6. Technical Highlights

6.1 Layered LLM Pipeline

STEP 1 (Event Understanding) ──► STEP 2 (Transmission Analysis) ──► STEP 3 (Signal Generation)

Each step's output feeds into the next, with context progressively enriching. All LLM calls require structured JSON output with Pydantic validation for type safety.

6.2 Knowledge Graph as Deterministic Skeleton

# Industry Knowledge Graph (networkx + JSON)
# Nodes: Crude Oil, Natural Gas, Petrochem, Aviation...
# Edges: Crude Oil → Petrochem (cost transmission, 70%, 7 days)
# Edges: Crude Oil → Marine (cost transmission, 80%, 1 day)

# During transmission analysis, inject formatted graph constraints
# "Prioritize graph constraints, supplement cross-industry links"

Knowledge graph provides deterministic industry chain relationships; LLM reasons beyond coverage for cross-industry generalization while maintaining accuracy.

6.3 Dynamic Learning Feedback

graph LR
    A[Analysis Result] --> B[Dynamic Learning]
    B --> C{New Relationship?}
    C -->|YES| D[evidence_count++]
    D --> E{Confidence > Threshold?}
    E -->|YES| F[Merge to Graph]
    F --> G[Enriched Knowledge Graph]
    G --> A
    C -->|NO| H[Skip]
    E -->|NO| H
Loading

From analysis results, automatically learn new industry chain relationships (evidence_count increments to boost confidence), discovered event patterns are used for type matching enhancement.

6.4 Backtest-Driven Confidence Calibration

Current Event ──► Historical Similarity Match ──► Confidence Adjustment
                  (Weighted: Title 25%, Type 20%, Entity 30%, Sentiment 25%)

High Similarity (>0.8) ──► +10% Confidence
Low Similarity (<0.4)  ──► -10% Confidence

Not executing backtest trades, but dynamically adjusting signal confidence based on historical event patterns to improve signal reliability.

6.5 Event Type Driven RAG Context

EVENT_TYPE_CONTEXT_MAP = {
    "geo-political": {
        "news_priority": ["geo", "conflict", "sanction", "diplomacy"],
        "market_data": ["energy", "military", "gold", "forex"],
        "include_hot_sectors": True
    },
    "policy": {
        "news_priority": ["central bank", "fiscal", "regulation"],
        "market_data": ["bank", "securities", "real estate", "insurance"],
        ...
    },
    ...
}

RAG service selectively injects relevant context (news priority + market data types) based on event type, not retrieving all data to reduce noise.

6.6 Pure SVG Hand-Drawn Transmission Graph

Frontend implements pure SVG hand-drawn transmission network (no third-party libraries), featuring:

  • Layered Layout Algorithm (stratified by transmission depth, same layer vertically centered)
  • Bezier Curve Edges (cubic Bezier curves connecting nodes)
  • Edge Width Encoding (higher transmission rate = thicker edge)
  • Interactive Hover (highlight related edges, display Tooltip)

7. System Demonstration

7.1 Input

US-IRAN military conflict, Iran blockades Strait of Hormuz

7.2 Output

Step 1: Event Understanding

  • Event Type: Geo-political
  • Core Entities: USA, Iran, Strait of Hormuz
  • Market Sentiment: -0.85 (extremely bearish)
  • Direct Impact: Crude Oil (supply reduction, +8), Marine (cost increase, -5)

Step 2: Transmission Analysis

Source Target Type Rate Delay
CRUDE OIL REFINING Cost Transmission 70% 3d
CRUDE OIL PETROCHEM Cost Transmission 60% 7d
CRUDE OIL MARINE Cost Transmission 80% 1d
MARINE AVIATION Cost Transmission 70% 5d

Step 3: Trading Signals

Industry Signal Confidence Related Assets
Oil Exploration LONG 85% CNOOC, XOM, CVX
Marine Shipping SHORT 75% Cosco, CMA CGM
Aviation SHORT 70% Air China, American Airlines

8. Tech Stack

Layer Technology
Backend FastAPI, Python 3.11+, Pydantic, networkx, uvicorn
LLM SiliconFlow (Recommended), OpenAI, Claude, Ollama
Frontend Next.js 14, React 18, TypeScript, Tailwind CSS
Graph networkx, pyvis (Interactive), Pure SVG (Hand-drawn)
Data AkShare, Tushare (Free Data Sources)
Infra Docker, Docker Compose v2, Nginx

9. Project Structure

event-trading-system/
├── deploy.sh              # VPS one-click deployment script
├── docker-compose.yml     # Container orchestration (with Nginx)
├── nginx.conf             # Nginx configuration (HTTP + HTTPS template)
├── .env.example           # Environment variables template
│
├── backend/
│   ├── app/
│   │   ├── main.py           # FastAPI entry point
│   │   ├── config.py         # Configuration management
│   │   ├── schemas/
│   │   │   └── models.py     # Pydantic data models
│   │   └── services/
│   │       ├── llm_client.py       # Multi-LLM wrapper + fallback
│   │       ├── prompts.py          # Prompt templates
│   │       ├── analysis.py         # Event analysis (core orchestration)
│   │       ├── rag_service.py      # RAG real-time context
│   │       ├── network_graph.py    # Transmission graph generation
│   │       ├── semantic_matcher.py # Semantic matcher
│   │       ├── causal_reasoning/   # Causal reasoning module
│   │       │   ├── industry_graph.py
│   │       │   ├── dynamic_graph_learner.py
│   │       │   └── llm_reasoner.py
│   │       ├── signal_output/
│   │       │   └── llm_explainer.py
│   │       ├── impact_quant/       # Backtest system
│   │       │   ├── event_backtest.py
│   │       │   └── historical_signals.py
│   │       ├── event_extraction/
│   │       │   └── llm_extractor.py
│   │       └── data_providers/      # Data providers
│   │           ├── base.py
│   │           ├── akshare_provider.py
│   │           └── tushare_provider.py
│   └── requirements.txt
│
└── frontend/
    ├── src/
    │   ├── app/page.tsx
    │   ├── components/
    │   │   ├── TransmissionGraph.tsx
    │   │   ├── TransmissionChain.tsx
    │   │   ├── SentimentMeter.tsx
    │   │   ├── DegradationBanner.tsx
    │   │   └── Sidebar.tsx
    │   ├── hooks/
    │   │   └── useAnalysis.ts
    │   └── types/
    │       └── index.ts
    └── Dockerfile

10. API Reference

10.1 Analyze Event (Non-Streaming)

POST /api/v1/analyze
Content-Type: application/json

{
  "title": "US-IRAN Military Conflict",
  "content": "Iran announces blockade of Strait of Hormuz..."
}

10.2 Analyze Event (Streaming)

POST /api/v1/analyze/stream

Response: Server-Sent Events (SSE) stream

10.3 Knowledge Graph

# Statistics
GET /api/v1/knowledge-graph/stats

# Relationships
GET /api/v1/knowledge-graph/relationships?min_confidence=0.5

10.4 Historical Events

GET /api/v1/history/events

10.5 Health Check

GET /health

11. Deployment Guide

11.1 Local Deployment

Prerequisites

  • Docker Desktop 20.10+
  • Git

Steps

# 1. Clone repository
git clone https://github.com/mkih76/event-driven-trading-system.git
cd event-driven-trading-system

# 2. Configure environment
cp .env.example .env
# Edit .env to add SILICONFLOW_API_KEY

# 3. Start services
docker compose up -d

# 4. Access
# Frontend: http://localhost
# API: http://localhost:8080
# Health: http://localhost:8080/health

Manual Setup (Alternative)

Backend:

cd backend
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp ../.env.example .env
# Edit .env to add API Key
uvicorn app.main:app --reload --port 8080

Frontend:

cd frontend
npm install
npm run dev
# Access http://localhost:3000

11.2 Cloud Deployment (VPS)

Prerequisites

  • Ubuntu 20.04+ / Debian 11+
  • 2G+ RAM (vector model loading needs 300-500MB)
  • Domain name (optional, for HTTPS)

One-Click Deployment

# Login to VPS as root
bash <(curl -sL https://raw.githubusercontent.com/mkih76/event-driven-trading-system/master/deploy.sh)

The script will:

  1. Install Docker and Docker Compose
  2. Clone/pull the repository
  3. Configure environment variables (interactive)
  4. Set up firewall (ports 22, 80, 443)
  5. Build and start containers
  6. Perform health checks

Manual VPS Setup

# 1. Install Docker
curl -fsSL https://get.docker.com | sh
systemctl enable docker

# 2. Install Docker Compose
curl -L "https://github.com/docker/compose/releases/latest/download/docker-compose-$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose
chmod +x /usr/local/bin/docker-compose

# 3. Create directory
mkdir -p /opt/event-trading-system
cd /opt/event-trading-system

# 4. Clone repository
git clone https://github.com/mkih76/event-driven-trading-system.git .

# 5. Configure environment
cp .env.example .env
# Edit .env with your API keys

# 6. Start services
docker compose up -d --build

# 7. Configure firewall
ufw allow 22/tcp
ufw allow 80/tcp
ufw allow 443/tcp
ufw --force enable

Optional: HTTPS Configuration

# Install certbot
apt-get update && apt-get install -y certbot python3-certbot-nginx

# Obtain SSL certificate
certbot --nginx -d your-domain.com --noninteractive --agree-tos -m admin@your-domain.com

# Auto-renewal (optional)
certbot renew --dry-run

Post-Deployment Management

Command Description
docker compose logs -f View real-time logs
docker compose restart Restart all services
docker compose restart backend Restart backend only
git pull && docker compose up -d --build Update and rebuild
docker compose down Stop all services
docker compose exec backend python -m pytest Run tests

Troubleshooting

# Check container status
docker compose ps

# Check logs
docker compose logs backend
docker compose logs frontend
docker compose logs nginx

# Rebuild specific service
docker compose up -d --build backend

# Reset database (if needed)
docker compose exec backend rm -f /app/data/events.db
docker compose restart backend

12. License

MIT License

Copyright (c) 2024 Event Trading Intelligence System

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

中文版 | Chinese Version

目录

  1. 系统架构
  2. 分析流程
  3. 传导推理示例
  4. 项目理念
  5. 核心能力
  6. 技术亮点
  7. 系统演示
  8. 技术栈
  9. 项目结构
  10. 接口文档
  11. 部署指南
  12. 许可证

1. 系统架构

┌─────────────────────────────────────────────────────────────────────────────────────────────┐
│                                        用户输入                                                │
│                              "美国与伊朗军事冲突"                                             │
└─────────────────────────────────────────────────────────────────────────────────────────────┘
                                              │
                    ┌─────────────────────────┼─────────────────────────┐
                    │                         │                         │
                    ▼                         ▼                         ▼
┌───────────────────────────────────┐ ┌───────────────────────────────┐ ┌───────────────────┐
│  ┌─────────┐ ┌─────────┐         │ │ ┌─────────┐ ┌─────────┐       │ │ ┌─────────┐       │
│  │ 事件    │ │ 传导    │         │ │ │ 传导    │ │ 历史    │       │ │ │ 信号    │       │
│  │ 摘要    │ │ 图谱    │         │ │ │ 明细    │ │ 回测    │       │ │ │ 报告    │       │
│  └────┬────┘ └────┬────┘         │ │ └────┬────┘ └────┬────┘       │ │ └────┬────┘       │
└───────┼───────────┼───────────────┼─┼──────┼──────────┼────────────┼─┼──────┼─────────────┘
        │           │               │ │      │          │            │ │      │
        │    SSE 流式响应 ◄────────┘ │      │          │            │ │      │
        │           │                 │      │          │            │ │      │
        ▼           ▼                 ▼      ▼          ▼            ▼ ▼      ▼
┌─────────────────────────────────────────────────────────────────────────────────────────────┐
│                                        FastAPI 网关                                            │
│                           /api/v1/analyze  |  /api/v1/analyze/stream                           │
└─────────────────────────────────────────────────────────────────────────────────────────────┘
                                              │
                    ┌─────────────────────────┼─────────────────────────┐
                    │                         │                         │
                    ▼                         ▼                         ▼
        ┌───────────────────┐     ┌───────────────────────┐     ┌───────────────────┐
        │      步骤 1        │     │       步骤 2           │     │      步骤 3       │
        │  ┌─────────────┐  │     │  ┌─────────────────┐  │     │  ┌─────────────┐  │
        │  │ 事件理解    │  │     │  │ 传导分析        │  │     │  │ 信号生成    │  │
        │  │   (LLM)    │  │───► │  │ (LLM + 图谱)   │  │───► │  │   (LLM)    │  │
        │  └─────────────┘  │     │  └────────┬────────┘  │     │  └─────────────┘  │
        └───────────────────┘     └───────────┼───────────┘     └───────────────────┘
                                              │
                    ┌─────────────────────────┼─────────────────────────┐
                    │                         │                         │
                    ▼                         ▼                         ▼
           ┌────────────────┐      ┌────────────────────┐       ┌────────────────┐
           │    知识图谱     │      │      动态学习        │       │    回测引擎   │
           │  产业链关系     │      │    反馈闭环         │       │  置信度校准    │
           └────────┬───────┘      └─────────┬──────────┘       └────────┬───────┘
                    │                         │                         │
                    └─────────────────────────┼─────────────────────────┘
                                              │
                                              ▼
                                 ┌────────────────────────┐
                                 │      RAG 服务          │
                                 │  实时市场数据注入       │
                                 └───────────┬────────────┘
                                             │
                                             ▼
                                 ┌────────────────────────┐
                                 │   AkShare / Tushare    │
                                 │   财经数据源           │
                                 └────────────────────────┘

2. 分析流程

sequenceDiagram
    autonumber
    participant User as 用户
    participant Frontend as 前端
    participant API as FastAPI
    participant LLM as LLM
    participant KG as 知识图谱
    participant BT as 回测引擎
    participant DL as 动态学习

    User->>Frontend: 提交事件
    Frontend->>API: POST /analyze/stream
    API->>LLM: 步骤1: 事件理解
    Note over LLM: 识别类型、实体、情绪

    API->>LLM: 步骤2: 传导分析
    LLM->>KG: 查询产业链约束
    KG-->>LLM: 返回关系路径

    API->>LLM: 步骤3: 信号生成
    API->>BT: 历史相似度匹配
    BT-->>API: 置信度调整

    API-->>Frontend: SSE 流式响应
    Frontend-->>User: 实时可视化

    API->>DL: 学习新关系
    DL-->>KG: 更新知识图谱
Loading

3. 传导推理示例

3.1 流程图

graph TB
    subgraph Event["地缘政治事件"]
        IRAN["伊朗封锁霍尔木兹海峡"]
    end

    subgraph Supply["供给冲击"]
        OIL["原油 供给 ↓15%"]
    end

    subgraph Transmission["产业链传导"]
        REFIN["炼化 成本 ↑20%"]
        CHEM["化工 成本 ↑15%"]
        SHIP["航运 成本 ↑25%"]
    end

    subgraph Demand["需求影响"]
        AVIATION["航空 需求 ↓10%"]
        TOURISM["旅游 需求 ↓8%"]
    end

    subgraph Safe["避险资产"]
        GOLD["黄金 ↑5-8%"]
    end

    IRAN-->OIL
    OIL-->REFIN
    OIL-->CHEM
    OIL-->SHIP
    SHIP-->AVIATION
    OIL-.->TOURISM
    OIL-.->GOLD
Loading

3.2 传导链路

路径 类型 传导率 延迟
伊朗 → 原油 供给冲击 100% 1-3天
原油 → 炼化 成本传导 70% 3天
原油 → 化工 成本传导 60% 7天
原油 → 航运 成本传导 80% 1天
航运 → 航空 成本传导 70% 5天
原油 → 旅游 风险偏好 - -
原油 → 黄金 避险需求 - -

4. 项目理念

在全球化市场中，重大地缘政治事件、政策变化、突发灾难往往在数小时至数天内沿产业链快速传导，引发相关行业股价的连锁反应。传统量化模型依赖历史数据，难以捕捉这些「黑天鹅」事件的非典型影响。

本系统利用大语言模型的逻辑推理能力，从事件原文出发，自动构建产业链传导路径，并结合知识图谱约束、回测验证、实时市场数据，生成可执行的交易信号。

核心创新

创新点 说明
LLM 驱动推理 超越历史模式推理新颖事件
图谱约束传导 确定性关系 + 灵活泛化
置信度校准 回测驱动的信号可靠性
自学习闭环 持续的知识图谱扩充

5. 核心能力

能力 说明
事件理解 LLM 自动识别事件类型（地缘政治/政策/灾难/经济数据）、核心实体、市场情绪
产业链传导 多级传导推理（成本传导 / 需求传导 / 替代效应）
交易信号 做多/做空信号及置信度，支持 A 股、美股、ETF
回测验证 基于历史相似事件动态调整置信度
动态学习 自动发现新产业链关系，构建反馈闭环
实时流式 SSE 流式展示 LLM 推理过程

6. 技术亮点

6.1 分层递进式 LLM Pipeline

步骤 1 (事件理解) ──► 步骤 2 (传导分析) ──► 步骤 3 (信号生成)

每一步的输出作为下一步的输入，上下文逐步丰富。所有 LLM 调用要求结构化 JSON 输出，Pydantic 验证确保类型安全。

6.2 知识图谱作为确定性骨架

# 产业链知识图谱 (networkx + JSON)
# 节点: 原油、天然气、化工、航空...
# 边: 原油→化工 (成本传导, 70%, 7天)
# 边: 原油→航运 (成本传导, 80%, 1天)

# 传导分析时，注入格式化图谱约束
# "优先遵循图谱约束，补充跨行业联系"

知识图谱提供确定性产业链关系，LLM 在其基础上推理未覆盖的跨行业联系，平衡泛化能力与准确性。

6.3 动态学习反馈闭环

graph LR
    A[分析结果] --> B[动态学习]
    B --> C{新关系?}
    C -->|是| D[evidence_count++]
    D --> E{置信度>阈值?}
    E -->|是| F[合并到图谱]
    F --> G[丰富的图谱]
    G --> A
    C -->|否| H[跳过]
    E -->|否| H
Loading

从分析结果中自动学习新产业链关系（evidence_count 增量提升置信度），发现的事件模式用于类型匹配增强。

6.4 回测驱动的置信度校准

当前事件 ──► 历史相似事件匹配 ──► 置信度调整
           (加权: 标题25%, 类型20%, 实体30%, 情绪25%)

高相似度(>0.8) ──► +10% 置信度
低相似度(<0.4)  ──► -10% 置信度

不是执行回测交易，而是基于历史事件模式动态调整信号置信度，提升信号可靠性。

6.5 事件类型驱动的 RAG 上下文

EVENT_TYPE_CONTEXT_MAP = {
    "地缘政治": {
        "news_priority": ["地缘", "冲突", "制裁", "外交"],
        "market_data": ["能源", "军工", "黄金", "外汇"],
        "include_hot_sectors": True
    },
    "政策": {
        "news_priority": ["央行", "财政", "监管", "政策"],
        "market_data": ["银行", "证券", "房地产", "保险"],
        ...
    },
    ...
}

RAG 服务根据事件类型选择性注入相关上下文（新闻优先级 + 市场数据类型），而非检索全部数据，降低噪声。

6.6 纯 SVG 手绘传导图谱

前端使用纯 SVG 手绘传导网络图（非第三方图库），实现了：

  • 分层布局算法（按传导深度分层，同层垂直居中）
  • 贝塞尔曲线边（三阶贝塞尔曲线连接节点）
  • 边宽度编码（传导率越高边越粗）
  • 交互悬停（高亮相关边，显示 Tooltip）

7. 系统演示

7.1 输入

美国与伊朗军事冲突，伊朗封锁霍尔木兹海峡

7.2 输出

步骤 1: 事件理解

  • 事件类型: 地缘政治
  • 核心实体: 美国、伊朗、霍尔木兹海峡
  • 市场情绪: -0.85（极度利空）
  • 直接影响: 原油（供给减少，+8）、航运（成本上升，-5）

步骤 2: 传导分析

来源 目标 类型 传导率 延迟
原油 炼化 成本传导 70% 3天
原油 化工 成本传导 60% 7天
原油 航运 成本传导 80% 1天
航运 航空 成本传导 70% 5天

步骤 3: 交易信号

行业 信号 置信度 相关标的
石油开采 做多 85% 中海油、中国石油、XOM
航运 做空 75% 中远海控、达飞海运
航空 做空 70% 中国国航、美国航空

8. 技术栈

层级 技术
后端 FastAPI, Python 3.11+, Pydantic, networkx, uvicorn
LLM SiliconFlow（推荐）、OpenAI、Claude、Ollama
前端 Next.js 14, React 18, TypeScript, Tailwind CSS
图谱 networkx, pyvis（交互图）、纯 SVG（手绘图）
数据 AkShare, Tushare（免费数据源）
容器 Docker, Docker Compose v2, Nginx

9. 项目结构

event-trading-system/
├── deploy.sh              # VPS 一键部署脚本
├── docker-compose.yml     # 容器编排（含 Nginx）
├── nginx.conf             # Nginx 配置（HTTP + HTTPS 模板）
├── .env.example           # 环境变量模板
│
├── backend/
│   ├── app/
│   │   ├── main.py           # FastAPI 入口
│   │   ├── config.py         # 配置管理
│   │   ├── schemas/
│   │   │   └── models.py     # Pydantic 数据模型
│   │   └── services/
│   │       ├── llm_client.py       # 多 LLM 封装 + 降级
│   │       ├── prompts.py          # Prompt 模板
│   │       ├── analysis.py         # 事件分析（核心编排）
│   │       ├── rag_service.py      # RAG 实时上下文
│   │       ├── network_graph.py    # 传导图谱生成
│   │       ├── semantic_matcher.py # 语义匹配器
│   │       ├── causal_reasoning/   # 因果推理模块
│   │       │   ├── industry_graph.py
│   │       │   ├── dynamic_graph_learner.py
│   │       │   └── llm_reasoner.py
│   │       ├── signal_output/
│   │       │   └── llm_explainer.py
│   │       ├── impact_quant/       # 回测系统
│   │       │   ├── event_backtest.py
│   │       │   └── historical_signals.py
│   │       ├── event_extraction/
│   │       │   └── llm_extractor.py
│   │       └── data_providers/      # 数据提供者
│   │           ├── base.py
│   │           ├── akshare_provider.py
│   │           └── tushare_provider.py
│   └── requirements.txt
│
└── frontend/
    ├── src/
    │   ├── app/page.tsx
    │   ├── components/
    │   │   ├── TransmissionGraph.tsx
    │   │   ├── TransmissionChain.tsx
    │   │   ├── SentimentMeter.tsx
    │   │   ├── DegradationBanner.tsx
    │   │   └── Sidebar.tsx
    │   ├── hooks/
    │   │   └── useAnalysis.ts
    │   └── types/
    │       └── index.ts
    └── Dockerfile

10. 接口文档

10.1 分析事件（非流式）

POST /api/v1/analyze
Content-Type: application/json

{
  "title": "美国与伊朗军事冲突",
  "content": "伊朗宣布封锁霍尔木兹海峡..."
}

10.2 分析事件（流式）

POST /api/v1/analyze/stream

响应: Server-Sent Events (SSE) 流

10.3 知识图谱

# 统计信息
GET /api/v1/knowledge-graph/stats

# 关系查询
GET /api/v1/knowledge-graph/relationships?min_confidence=0.5

10.4 历史事件

GET /api/v1/history/events

10.5 健康检查

GET /health

11. 部署指南

11.1 本地部署

前置条件

  • Docker Desktop 20.10+
  • Git

步骤

# 1. 克隆仓库
git clone https://github.com/mkih76/event-driven-trading-system.git
cd event-driven-trading-system

# 2. 配置环境变量
cp .env.example .env
# 编辑 .env 添加 SILICONFLOW_API_KEY

# 3. 启动服务
docker compose up -d

# 4. 访问
# 前端: http://localhost
# API: http://localhost:8080
# 健康检查: http://localhost:8080/health

手动部署（备选）

后端：

cd backend
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp ../.env.example .env
# 编辑 .env 添加 API Key
uvicorn app.main:app --reload --port 8080

前端：

cd frontend
npm install
npm run dev
# 访问 http://localhost:3000

11.2 云服务器部署（VPS）

前置条件

  • Ubuntu 20.04+ / Debian 11+
  • 2G+ 内存（向量模型加载需要 300-500MB）
  • 域名（可选，用于 HTTPS）

一键部署

# 登录 VPS 并以 root 运行
bash <(curl -sL https://raw.githubusercontent.com/mkih76/event-driven-trading-system/master/deploy.sh)

脚本将执行以下操作：

  1. 安装 Docker 和 Docker Compose
  2. 克隆/更新代码仓库
  3. 配置环境变量（交互式）
  4. 设置防火墙（开放 22、80、443 端口）
  5. 构建并启动容器
  6. 执行健康检查

手动 VPS 部署

# 1. 安装 Docker
curl -fsSL https://get.docker.com | sh
systemctl enable docker

# 2. 安装 Docker Compose
curl -L "https://github.com/docker/compose/releases/latest/download/docker-compose-$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose
chmod +x /usr/local/bin/docker-compose

# 3. 创建目录
mkdir -p /opt/event-trading-system
cd /opt/event-trading-system

# 4. 克隆仓库
git clone https://github.com/mkih76/event-driven-trading-system.git .

# 5. 配置环境变量
cp .env.example .env
# 编辑 .env 填入 API Key

# 6. 启动服务
docker compose up -d --build

# 7. 配置防火墙
ufw allow 22/tcp
ufw allow 80/tcp
ufw allow 443/tcp
ufw --force enable

可选：HTTPS 配置

# 安装 certbot
apt-get update && apt-get install -y certbot python3-certbot-nginx

# 申请 SSL 证书
certbot --nginx -d your-domain.com --noninteractive --agree-tos -m admin@your-domain.com

# 自动续期（可选）
certbot renew --dry-run

部署后管理

命令 说明
docker compose logs -f 实时查看日志
docker compose restart 重启所有服务
docker compose restart backend 仅重启后端
git pull && docker compose up -d --build 更新并重新构建
docker compose down 停止所有服务
docker compose exec backend python -m pytest 运行测试

故障排查

# 查看容器状态
docker compose ps

# 查看日志
docker compose logs backend
docker compose logs frontend
docker compose logs nginx

# 重新构建指定服务
docker compose up -d --build backend

# 重置数据库（如需要）
docker compose exec backend rm -f /app/data/events.db
docker compose restart backend

12. 许可证

MIT 许可证

版权所有 (c) 2024 Event Trading Intelligence System

特此向任何获得本软件及相关文档文件（"软件"）副本的人免费授予无限制使用、复制、分发、再分发本软件的许可，包括无限制使用、复制、修改、合并、发布、分发、再许可和/或出售本软件副本的权利，允许向提供本软件的人这样做，但须满足以下条件：

上述版权声明和本许可声明应包含在本软件的所有副本或重要部分中。

本软件按"原样"提供，不提供任何明示或暗示的保证，包括但不限于对适销性、特定用途适用性和非侵权性的保证。在任何情况下，作者或版权持有人均不对因本软件或使用或其他与本软件相关的交易而产生的任何索赔、损害或其他责任负责，无论是在合同诉讼、侵权诉讼或其他诉讼中。

About

事件交易驱动系统

Topics

knowledge-graph event-driven causal-inference stock-analysis trading-servers

Resources

Readme
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages