Dataset download link: https://modelscope.cn/datasets/CMB_AILab/YiZhao-FinDataSet/summary

Project Structure (Hierarchy & File Descriptions)

Hierarchical_Multi_Agent/
├── config/                          # Configuration
│   ├── default.yaml                 # Default HMA and evidence-theory parameters
│   ├── experiment/
│   │   ├── hma_full.yaml            # Full HMA experiment
│   │   ├── ablation_scoring_only.yaml # Ablation: scoring agents only
│   │   └── baselines.yaml           # Baseline comparison experiments
│   └── env/
│       ├── development.yaml
│       └── production.yaml
├── src/
│   ├── core/                        # Core algorithms
│   │   ├── evidence_theory/         # Evidence theory
│   │   │   ├── bpa_generator.py     # BPA generation and normalization
│   │   │   ├── dempster_fusion.py   # Dempster’s combination rule
│   │   │   ├── conflict_measurement.py # Conflict measurement and weighted correction
│   │   │   └── uncertainty_analysis.py # Bel/Pl/evidence entropy
│   │   ├── framework/
│   │   │   ├── agent_registry.py     # Agent registry
│   │   │   └── hierarchical_agent_orchestrator.py # Hierarchical orchestration
│   │   └── risk_aware_feature_network/ # RAFN
│   │       ├── adaptive_attention.py
│   │       └── rafn_module.py
│   ├── agents/
│   │   ├── base/                    # Base classes and interfaces
│   │   ├── scoring/                 # Five types of scoring agents
│   │   ├── judging/                 # Judges and weighted calibration
│   │   └── reasoning/               # Reasoning agent and evidence fusion
│   ├── data/
│   │   ├── loaders/                 # YZF loading and jsonl parsing
│   │   ├── preprocessing/          # Text normalization and tokenization wrappers
│   │   └── knowledge_base/          # Local knowledge base and RAG
│   ├── models/
│   │   ├── baselines/               # TextCNN, BERT, RoBERTa, FinBERT
│   │   └── fusion/                  # Multi-source fusion interfaces
│   ├── evaluation/
│   │   ├── metrics/                 # MAE, MSE, RMSE, MAE/MSE, confidence intervals
│   │   └── visualization/          # Result plotting
│   └── utils/                       # Logging, checkpoints, dynamic weights
├── scripts/
│   ├── run_experiments/             # Run HMA / ablations / baselines
│   ├── data_preparation/            # Build sample data and knowledge base
│   └── evaluation/                 # Metric calculation and figure generation
├── tests/
│   ├── unit/                        # Evidence theory, agents, RAFN
│   └── integration/                # End-to-end HMA pipeline
├── resources/
│   ├── datasets/yzf_sample/         # YZF sample data (jsonl)
│   ├── knowledge_base/
│   └── pretrained/
├── outputs/                          # Experiment outputs (logs, metrics, figures)
├── requirements.txt
└── README.md

Quick Start

1. Environment

cd Hierarchical_Multi_Agent
pip install -r requirements.txt

2. Data Preparation

Dataset download link: https://modelscope.cn/datasets/CMB_AILab/YiZhao-FinDataSet/summary

python scripts/data_preparation/download_yzf_sample.py --output_dir resources/datasets/yzf_sample --num_en 100 --num_zh 100

3. Run the Full HMA Experiment

python scripts/run_experiments/run_hma_full.py --data_dir resources/datasets/yzf_sample --max_samples 100 --output_dir outputs/hma_full

4. Ablation Experiment (Scoring Agents Only)

python scripts/run_experiments/run_ablation.py --data_dir resources/datasets/yzf_sample --max_samples 200

5. Baseline Comparisons

python scripts/run_experiments/run_baselines.py --data_dir resources/datasets/yzf_sample --max_samples 300

6. Generate Comparison Figures

python scripts/evaluation/generate_figures.py --output outputs/figures/comparison.png --title "Risk assessment score comparison"

7. Run Tests

pip install pytest
pytest tests/ -v

Data Description

• The full YZF dataset is available at: ModelScope - YiZhao-Fin Data Set.
• The repository uses jsonl files under resources/datasets/yzf_sample, with fields including: id, url, title, source_domain, text, risk_score, fin_int_score, language, etc.

数据集下载地址: https://modelscope.cn/datasets/CMB_AILab/YiZhao-FinDataSet/summary

工程结构（层级与文件说明）

Hierarchical_Multi_Agent/
├── config/                          # 配置
│   ├── default.yaml                 # 默认 HMA 与证据理论参数
│   ├── experiment/
│   │   ├── hma_full.yaml            # 完整 HMA 实验
│   │   ├── ablation_scoring_only.yaml # 消融：仅评分智能体
│   │   └── baselines.yaml           # 基线对比实验
│   └── env/
│       ├── development.yaml
│       └── production.yaml
├── src/
│   ├── core/                        # 核心算法
│   │   ├── evidence_theory/         # 证据理论
│   │   │   ├── bpa_generator.py     # BPA 生成与归一化
│   │   │   ├── dempster_fusion.py   # Dempster 组合规则
│   │   │   ├── conflict_measurement.py # 冲突度量与加权修正
│   │   │   └── uncertainty_analysis.py # Bel/Pl/证据熵
│   │   ├── framework/
│   │   │   ├── agent_registry.py     # 智能体注册
│   │   │   └── hierarchical_agent_orchestrator.py # 层级编排
│   │   └── risk_aware_feature_network/ # RAFN
│   │       ├── adaptive_attention.py
│   │       └── rafn_module.py
│   ├── agents/
│   │   ├── base/                    # 基类与接口
│   │   ├── scoring/                 # 五类评分智能体
│   │   ├── judging/                 # 裁判与加权校准
│   │   └── reasoning/               # 推理智能体与证据融合
│   ├── data/
│   │   ├── loaders/                 # YZF 加载与 jsonl 解析
│   │   ├── preprocessing/          # 文本规范化与分词封装
│   │   └── knowledge_base/          # 本地知识库与 RAG
│   ├── models/
│   │   ├── baselines/               # TextCNN, BERT, RoBERTa, FinBERT
│   │   └── fusion/                  # 多源融合接口
│   ├── evaluation/
│   │   ├── metrics/                 # MAE, MSE, RMSE, MAE/MSE, 置信区间
│   │   └── visualization/          # 结果作图
│   └── utils/                       # 日志、 checkpoint、动态权重
├── scripts/
│   ├── run_experiments/             # 运行 HMA / 消融 / 基线
│   ├── data_preparation/            # 构造样本数据与知识库
│   └── evaluation/                 # 指标计算与图表生成
├── tests/
│   ├── unit/                        # 证据理论、智能体、RAFN
│   └── integration/                # 全流程 HMA
├── resources/
│   ├── datasets/yzf_sample/         # YZF 样本数据（jsonl）
│   ├── knowledge_base/
│   └── pretrained/
├── outputs/                          # 实验输出（日志、指标、图）
├── requirements.txt
└── README.md

快速开始

1. 环境

cd Hierarchical_Multi_Agent
pip install -r requirements.txt

2. 处理数据

Download address of the dataset: https://modelscope.cn/datasets/CMB_AILab/YiZhao-FinDataSet/summary

python scripts/data_preparation/download_yzf_sample.py --output_dir resources/datasets/yzf_sample --num_en 100 --num_zh 100

3. 运行完整 HMA 实验

python scripts/run_experiments/run_hma_full.py --data_dir resources/datasets/yzf_sample --max_samples 100 --output_dir outputs/hma_full

4. 消融实验（仅评分智能体）

python scripts/run_experiments/run_ablation.py --data_dir resources/datasets/yzf_sample --max_samples 200

5. 基线对比

python scripts/run_experiments/run_baselines.py --data_dir resources/datasets/yzf_sample --max_samples 300

6. 生成对比图

python scripts/evaluation/generate_figures.py --output outputs/figures/comparison.png --title "Risk assessment score comparison"

7. 运行测试

pip install pytest
pytest tests/ -v

数据说明

• 完整 YZF 数据集见 ModelScope - YiZhao-Fin Data Set。
• 使用 resources/datasets/yzf_sample 下的 jsonl，字段包含：id, url, title, source_domain, text, risk_score, fin_int_score, language 等。

License

与论文及数据集许可保持一致；代码仅供研究与复现使用。