Add comprehensive dataset card for FinGAIA

README.md

@@ -0,0 +1,104 @@
+---
+license: mit
+task_categories:
+  - image-text-to-text
+  - audio-to-text
+language:
+  - zh
+tags:
+  - finance
+  - agent
+  - benchmark
+  - multimodal
+  - tool-use
+  - code
+  - chinese
+---
+
+# FinGAIA: An End-to-End Benchmark for Evaluating AI Agents in Finance
+
+[Paper](https://huggingface.co/papers/2507.17186) | [GitHub](https://github.com/SUFE-AIFLM-Lab/FinGAIA)
+
+## Introduction
+The booming development of AI agents presents unprecedented opportunities for automating complex tasks across various domains. However, their multi-step, multi-tool collaboration capabilities in the financial sector remain underexplored. This paper introduces FinGAIA, an end-to-end benchmark designed to evaluate the practical abilities of AI agents in the financial domain.
+
+FinGAIA comprises 407 meticulously crafted tasks, spanning seven major financial sub-domains: securities, funds, banking, insurance, futures, trusts, and asset management. These tasks are organized into three hierarchical levels of scenario depth: basic business analysis, asset decision support, and strategic risk management. Our work provides the first agent benchmark closely related to the financial domain, aiming to objectively assess and promote the development of agents in this crucial field.
+
+## Dataset Structure and Scenarios
+FinGAIA includes 407 tasks, covering seven major financial sub-domains. The tasks are designed under the guidance of financial domain experts and constructed using real-world financial data to ensure their authenticity and relevance. They are organized into three hierarchical levels of scenario depth:
+
+| Scenario Depth        | Financial Scenario                  | Questions |
+|-----------------------|-------------------------------------|-----------|
+| Operational Analytics |   Customer Data Analytics           | 47        |
+|                       |   Transaction Risk Assessment       | 42        |
+|                       |   All                               | 89        |
+| Asset Decision        |   Financial Data Statistics         | 101       |
+|                       |   Loan Credit Analysis              | 43        |
+|                       |   Fraud Detection Analysis          | 41        |
+|                       |   All                               | 185       |
+| Strategic Risk        |   Risk Management Analysis          | 42        |
+|                       |   Portfolio Fund Allocation         | 40        |
+|                       |   Market Trend Forecasting          | 51        |
+|                       |   All                               | 133       |
+| **FinGAIA**           | **All**                             | **407**   |
+
+*   **Level 1: Basic Business Analysis**: Focuses on evaluating the agent's ability to handle fundamental financial knowledge and process multi-modal financial information.
+*   **Level 2: Asset Decision Support**: Concentrates on information integration, logical reasoning, and tool utilization for tasks of medium complexity.
+*   **Level 3: Strategic Risk Management**: Addresses high-complexity tasks that require multi-tool coordination and strategic planning, comprehensively assessing the model's reasoning ability and professionalism in real financial contexts.
+
+## Core Capabilities Assessed
+Solving FinGAIA tasks requires AI agents to possess and coordinate multiple underlying capabilities, as detailed below:
+
+*   **Web Browsing Capability (42.7%)**: Searching and browsing websites to obtain real-time information.
+*   **File Processing Capability (18.1%)**: Handling various document formats such as PDF, CSV, and XLSX.
+*   **Multimodal Capability (15.2%)**: Understanding and analyzing non-textual data like images, charts, audio, and video.
+*   **Code Capability (12.2%)**: Executing Python and other code to perform data analysis and model calculations.
+*   **Computational Capability (10.1%)**: Performing algebraic operations.
+
+## Example Task
+Here's an example of a task from the "Customer Data Analysis" scenario, demonstrating the multi-modal and multi-step nature of the FinGAIA benchmark:
+
+**问题 (Question):**
+识别附件中的图片，这是一家期货公司的图标。访问其官网。在"营业机构模块"搜索：河北省沧州市营业网点的负责人是谁？
+
+*(English Translation: Identify the image in the attachment, which is the icon of a futures company. Visit its official website. In the "Branch Network" section, search for: Who is the person in charge of the Cangzhou branch in Hebei Province?)*
+
+**答案 (Answer):**
+冷俊杰
+
+*(English Translation: Leng Junjie)*
+
+**解题步骤 (Solution Steps):**
+1、识别附件图片 (1. Identify the attached image)
+2、辨别图片为中信建投期货图标 (2. Recognize the image as the CITIC Futures icon)
+3、访问官网 (3. Visit the official website)
+4、进入营业机构模块 (4. Navigate to the "Branch Network" module)
+5、筛选目标信息 (5. Filter for target information)
+6、查找负责人名称。(6. Find the name of the person in charge.)
+
+## Performance Evaluation
+The FinGAIA benchmark has been used to evaluate 10 mainstream AI agents in a zero-shot setting. The best-performing agent, ChatGPT, achieved an overall accuracy of 48.9%. While this is superior to non-professionals, it still lags financial experts by over 35 percentage points. Error analysis has revealed common failure patterns such as Cross-modal Alignment Deficiency, Financial Terminological Bias, and Operational Process Awareness Barrier, highlighting crucial directions for future research.
+
+The evaluation results indicate that while existing agents show promise in basic and medium-complexity tasks, significant room for improvement remains, especially in complex, multi-step reasoning and understanding tasks that require nuanced financial expertise and multi-tool coordination.
+
+## Usage
+You can download the FinGAIA dataset using `git lfs`. First, ensure Git LFS is installed, then clone the repository:
+
+```bash
+git lfs install
+git clone https://huggingface.co/datasets/SUFE-AIFLM-Lab/FinGAIA
+```
+
+For more detailed usage instructions on how to query models and use the benchmark, please refer to the [official GitHub repository](https://github.com/SUFE-AIFLM-Lab/FinGAIA).
+
+## Citation
+If you use FinGAIA in your research, please cite our paper:
+
+```bibtex
+@article{Fin-GAIA,
+  title={Fin-GAIA：一个用于评估金融领域 AI 代理的基准测试},
+  author={Fin-GAIA Team},
+  year={2024},
+  journal={arXiv preprint}
+}
+```