openhands-feedback

# OpenHands Feedback Dataset 🙌

## Dataset Description

### What is OpenHands Feedback?

The OpenHands Feedback Dataset is a collection of user interactions and feedback with the [OpenHands](https://github.com/OpenHands/OpenHands) AI coding assistant. This dataset contains real-world examples of how users interact with AI coding assistants, including both successful and unsuccessful interactions, along with user feedback on the quality and helpfulness of the responses.

The dataset currently contains 275 examples of user interactions (60% positive feedback and 40% negative), with a total size of approximately 310MB. Each example includes the complete interaction trajectory between the user and the assistant, along with the user's feedback and metadata about the interaction.

### Who Collects the Data?

The data is collected and stored by [All Hands AI](https://all-hands.dev), a company founded by OpenHands maintainers to support and improve OpenHands. When users provide feedback through the OpenHands interface (via thumbs-up/thumbs-down buttons), they can choose to submit their feedback either publicly or privately.

### Data Collection Process

Users can provide feedback at any point during their interaction with OpenHands by:
1. Clicking the thumbs-up or thumbs-down button
2. Providing their email address (optional, for potential follow-up questions)
3. Choosing whether to make their feedback public or private

For more details about the feedback collection process, see the [OpenHands Feedback Documentation](https://docs.all-hands.dev/modules/usage/feedback).

## Data Structure

The dataset is provided in Parquet format and contains the following fields:

- `version`: String identifying the version of the feedback format
- `feedback`: User feedback text
- `permissions`: Indicates whether the feedback is public or private
- `timestamp`: When the feedback was submitted
- `trajectory`: List of interaction steps containing:
- `content`: The content of the message
- `extras`: Additional metadata
- `id`: Message identifier
- `message`: The actual message text
- `observation`: System observations
- `source`: Message source (user/assistant)
- `timestamp`: When the message was sent

## Usage and Limitations

### Intended Uses

This dataset is intended to be used for:
- Improving AI coding assistants
- Studying human-AI interaction patterns
- Analyzing common failure modes in code generation
- Developing better evaluation metrics for coding assistants

### Privacy and Ethics

- Only publicly shared feedback is included in this dataset
- Personal information and email addresses have been removed
- The data has been reviewed to remove any sensitive information
- Users who contributed public feedback were given the opportunity to opt out before release

## Dataset Creation

### Data Quality

The dataset undergoes quality control before release:
- Removal of low-quality feedback
- Removal of submitter email addresses
- Review for sensitive information
- Validation of data format and completeness

### Updates

The dataset will be updated periodically as new public feedback is collected and processed. Updates will follow the same quality control and privacy protection processes.

## Additional Information

### License
This dataset is released under the MIT License, the same license as OpenHands itself.

### Citation

If you use this dataset in your research, please cite it as:

```bibtex
@misc{openhands-feedback,
title={OpenHands Feedback Dataset},
author={All Hands AI},
year={2024},
howpublished={\url{https://huggingface.co/datasets/all-hands/openhands-feedback}}
}
```

### Feedback and Support

For questions, concerns, or to request data deletion:
- Contact: [e-mail]
- GitHub: [OpenHands Repository](https://github.com/OpenHands/OpenHands)

To request deletion of specific data:
1. For individual entries: Use the link and password provided during submission
2. For all your data: Contact [e-mail] from the email address used during submission

# OpenHands 反馈数据集 🙌

## 数据集描述

### 什么是OpenHands反馈？

OpenHands 反馈数据集（OpenHands Feedback Dataset）是与[OpenHands](https://github.com/OpenHands/OpenHands) AI编码助手相关的用户交互与反馈集合。该数据集收录了用户与AI编码助手交互的真实场景案例，涵盖成功与失败的交互场景，同时包含用户对助手回复质量与实用性的反馈。

当前数据集共包含275条用户交互案例（其中60%为正面反馈，40%为负面反馈），总大小约310MB。每条案例均包含用户与助手的完整交互轨迹，以及用户反馈与交互相关的元数据。

### 数据收集方

本数据集由[All Hands AI](https://all-hands.dev)收集并存储，该公司由OpenHands的维护者创立，旨在支持并改进OpenHands项目。当用户通过OpenHands界面（通过点赞/点踩按钮）提交反馈时，可以选择将反馈设为公开或私密。

### 数据收集流程

用户可在与OpenHands交互的任意节点通过以下方式提交反馈：
1. 点击点赞或点踩按钮
2. 提供电子邮箱（可选，用于后续跟进沟通）
3. 选择将反馈设为公开或私密

如需了解更多反馈收集流程细节，请参阅[OpenHands 反馈文档](https://docs.all-hands.dev/modules/usage/feedback)。

## 数据结构

数据集以Parquet格式提供，包含以下字段：
- `version`：用于标识反馈格式版本的字符串
- `feedback`：用户提交的反馈文本
- `permissions`：标识反馈为公开或私密的字段
- `timestamp`：反馈提交时间
- `trajectory`：包含交互步骤的列表，其中每个步骤包含：
- `content`：消息内容
- `extras`：附加元数据
- `id`：消息标识符
- `message`：实际消息文本
- `observation`：系统观测信息
- `source`：消息来源（用户/助手）
- `timestamp`：消息发送时间

## 使用场景与限制

### 预期用途
本数据集可用于以下场景：
- 改进AI编码助手性能
- 研究人机交互模式
- 分析代码生成中的常见失效模式
- 开发更优质的编码助手评估指标

### 隐私与伦理
- 本数据集仅收录公开共享的反馈内容
- 已移除所有个人信息与电子邮箱地址
- 已对数据进行审核，移除所有敏感信息
- 公开反馈的贡献者在数据集发布前已获得退选选择权

## 数据集制作

### 数据质量管控
本数据集在发布前会经过严格的质量管控流程：
- 移除低质量反馈
- 移除提交者的电子邮箱地址
- 审核并移除敏感信息
- 验证数据格式与完整性

### 更新机制
本数据集将定期更新，收录并处理新收集的公开反馈。更新流程将遵循相同的质量管控与隐私保护标准。

## 附加信息

### 许可证
本数据集采用MIT许可证发布，与OpenHands项目本身使用的许可证一致。

### 引用说明
若您在研究中使用本数据集，请按以下方式引用：
bibtex
@misc{openhands-feedback,
title={OpenHands 反馈数据集},
author={All Hands AI},
year={2024},
howpublished={url{https://huggingface.co/datasets/all-hands/openhands-feedback}}
}

### 反馈与支持
如需咨询问题、反馈顾虑或申请删除数据，请通过以下方式联系：
- 邮箱：[e-mail]
- GitHub：[OpenHands 仓库](https://github.com/OpenHands/OpenHands)

如需申请删除特定数据：
1. 针对单条数据：使用提交时提供的链接与密码
2. 针对所有数据：使用提交反馈时使用的电子邮箱发送邮件至[e-mail]