README.md

---
license: mit
task_categories:
- question-answering
- text-generation
language:
- en
tags:
- finance
---
# Financial Multimodal Mathematical Reasoning QA Dataset💰
[[🔗Github](https://github.com/biggestCjb/FinMR)] [[📖 ArXiv Paper(not publish yet)](https://arxiv.org)]


## 💻Data Usage
```python
from datasets import load_dataset

dataset = load_dataset("aminous1/FinMR", cache_dir="/your/custom/path")
```
## 👋😊✨Dataset Description

FinQA is a dataset designed for financial reasoning and question answering. It includes questions, 
financial contexts, and corresponding answers. The dataset contains both textual and visual data, with visual data represented by images linked .

<p align="center">
    <img src="https://img520.com/Wevhgc.png" width="60%">  <img src="https://i.postimg.cc/gc7WJtM7/Error-Database.png" width="66%">
  These two images illustrate the input sample and the processing flow, respectively
</p>
<br>

## 📁Dataset Structure

The dataset contains the following fields:

1. **`Question ID`**
   - **Purpose**: A unique identifier for the question, used for indexing and management.
   - **Example**: `0`

2. **`Share Context`**
   - **Purpose**: Provides background information related to the question, helping the user understand the context and logic.
   - **Example**: Describes Wakuluk's analytical approach and economic predictions.

3. **`Share Image`**
   - **Purpose**: Contains paths or links to images that supplement the question, such as graphs or tables.
   - **Example**:
      ```list
      ["images/CapitalMarketExpectations_images/share1-8_1.png", "images/FixedIncome_images/share1-9_2.png"]
     ```

4. **`Question Text`**
   - **Purpose**: The specific content of the question that the user is required to answer.
   - **Example**: `"Wakuluk most likely seeks to mitigate which of the following biases in developing capital market forecasts?"`

5. **`Image`**
   - **Purpose**: If the question relies on a standalone image, this field contains its path or link.
   - **Example**: `images/add_images/1.png` 

6. **`Options`**
   - **Purpose**: Provides multiple answer choices, typically labeled with letters (e.g., A, B, C).
   - **Example**:
     ```json
     {
       "A": "Availability",
       "B": "Time period",
       "C": "Survivorship"
     }
     ```

7. **`Answer`**
   - **Purpose**: Indicates the correct answer to the question.
   - **Example**: `"A"`

8. **`Explanation`**
   - **Purpose**: Provides a detailed explanation of the correct answer, helping users understand the reasoning behind it.
   - **Example**: Explains that Wakuluk avoids "Availability Bias" by relying on objective evidence and analytical procedures.

9. **`General Topics`**
   - **Purpose**: Categorizes the question into a specific topic or area of knowledge, making it easier to organize and group.
   - **Example**: `"Foundation of Risk Management"`

10. **`QA Type`**
    - **Purpose**: Specifies the type of question, such as knowledge-based reasoning or operational tasks.
    - **Example**: `"Knowledge Reasoning QA"`

11. **`Level of Difficulty`**
    - **Purpose**: Indicates the difficulty level of the question, helping users gauge the challenge.
    - **Example**: `"Easy"`

12. **`shared_description`**
    - **Purpose**: Describes any visual information (e.g., images or tables) that is relevant to the question, ensuring clarity for users.
    - **Example**: Summarizes the layout, title, and data in Exhibit 1.

13. **`description`**
    - **Purpose**: Further annotations or descriptions about the question, though this field is `null` in the current example.
    - **Example**: `This picture xxxx`

14. **`Datasplit`**
    - **Purpose**: Specifies the dataset partition the question belongs to, such as training, validation, or testing.
    - **Example**: `"train"`

15. **`Index`**
    - **Purpose**: The position of the question within the dataset, used for quick lookup.
    - **Example**: `0`



### 📝Annotations 

Annotations were performed by financial domain experts, ensuring high accuracy and consistency.

## ⚠️Bias, Risks, and Limitations

The dataset may contain biases inherent to the financial documents from which it was curated. Users should exercise caution when generalizing results.

## 💡Recommendations

Avoid overgeneralization of model outputs and consider domain-specific adaptations.

## 📑 Citation


## 🤝Authors' Organization 

## 📧Contact


# Still under maintenance...🔧✨