anyangsong's picture
Update README.md
983c2e1 verified
|
Raw History Blame Contribute Delete
7.09 kB
---
license: mit
gated: true
extra_gated_prompt: |
**Warning: This dataset contains sensitive content, including potentially explicit, violent, or politically sensitive material.**
**By requesting access, you acknowledge that:**
**- you are aware of the nature of this dataset;**
**- you will use it responsibly and for legitimate research purposes;**
**- you will comply with applicable laws and regulations.**
configs:
- config_name: text
default: true
data_files:
- split: test
path: preview/text-00000-of-00001.parquet
- config_name: image
data_files:
- split: test
path: preview/image-00000-of-00001.parquet
- config_name: audio
data_files:
- split: test
path: preview/audio-00000-of-00001.parquet
- config_name: unknown_image
data_files:
- split: test
path: preview/unknown_image-00000-of-00001.parquet
task_categories:
- text-classification
- image-classification
- audio-classification
- video-classification
- text-generation
language:
- zh
- en
size_categories:
- 1K<n<10K
tags:
- sensitive-content
- sexual-not-safe-for-work
- politically-sensitive
- violence
- profanity
---
# Blockchain-Sensitive-Detect-Data
[English README](README_EN.md)
复旦大学附属儿科医院-区块链敏感信息检测项目的多模态完整测试数据集。
项目仓库:https://github.com/anyangsong/Blockchain-Sensitive-Detect
数据集:https://huggingface.co/datasets/anyangsong/Blockchain-Sensitive-Detect-Data
checkpoints:https://huggingface.co/anyangsong/Blockchain-Sensitive-Detect-Checkpoints
数据以原始文件夹组织,覆盖文本、音频、图像与视频等样本。
该仓库不提供统一的 CSV、Parquet 或 JSONL 清单;类别信息主要由目录名和文件名携带。
> **内容警告:** 数据集包含辱骂、性内容、暴力、政治相关内容、误导性医疗信息、欺诈信息。使用者应仅在具备适当访问控制、伦理审查和当地法律依据的环境中处理这些内容。
>
> **本数据集全部样本仅面向多模态敏感信息检测的学术研究,部分样本本身属于违反中国当地法律法规的内容,仅用于算法研究测评,严禁直接传播、扩散样本原始内容。如您发现存在违反中国法律法规、侵害合法权益、不适宜公开传播的内容,请联系项目团队申请删除(负责人穆欣雅(邮箱:imvl@fudan.edu.cn),教授冯瑞(邮箱:fengrui@fudan.edu.cn)、教授张玥杰(邮箱:yjzhang@fudan.edu.cn))**
## 数据概览
| 模态 | 位置 | 样本数 |
| :-- | :-: | :--: |
| 文本 | `texts/text.json` | 235 条 |
| 图像 | `images/` | 318 条 |
| 音频 | `audios/` | 426 条 |
| 视频 | `videos/` | 200 条 |
| 其他数据 | 未知敏感图像:`unknown_images/`;机器生成视频:`machine_videos/avlip/` | 若干 |
## 目录结构
```text
.
├── texts/
│ └── text.json
├── audios/
│ ├── politic/
│ ├── profanity/
│ ├── safe/
│ ├── sexual/
│ └── violence/
├── images/
│ ├── other/
│ ├── personal/
│ ├── politic/
│ └── violence/
├── unknown_images/
│ ├── non-unknown-sensitive/
│ └── unknown-sensitive/
└── videos/
```
## 文本数据
`texts/text.json` 是一个 JSON 对象,一级键为 `敏感信息` 与 `非敏感信息`。每个一级类别下包含细分类。
| 类别 | 样本数 |
| ------------------------- | ------ |
| 敏感-个人属性(PII) | 25 |
| 敏感-虚假或误导性医疗信息 | 40 |
| 敏感-欺诈和诈骗信息 | 40 |
| 敏感-辱骂攻击 | 50 |
| 非敏感-个人属性 | 20 |
| 非敏感-健康状况 | 40 |
| 非敏感-医疗应用 | 20 |
| 合计 | 235 |
## 图像数据
图像数据分为敏感与非敏感两大类,其中敏感图像进一步细分为政治内容、个人隐私和暴力内容三类。
| 类别 | 类名 | 样本数 |
| ---- | ------------------------------ | ------ |
| 0 | 政治内容(Political) | 27 |
| 1 | 个人隐私(Personal Information) | 203 |
| 2 | 暴力内容(Violence) | 17 |
| 3 | 非敏感 | 71 |
| 合计 | — | 318 |
## 音频数据
音频数据分为四个敏感内容类别和一个正常内容类别。文件类型以 WAV 为主,包含少量 MP3。
| **类别** | **类名** | **样本数** |
| -------- | ------------------- | ---------- |
| 0 | 正常(Safe) | 226 |
| 1 | 政治敏感(Politic) | 50 |
| 2 | 暴力煽动(Violence) | 50 |
| 3 | 色情低俗(Sexual) | 50 |
| 4 | 仇恨谩骂(Profanity) | 50 |
| **合计** | — | 426 |
## 视频数据
视频数据分为六个敏感内容类别和一个正常内容类别,源于被广泛用于弱监督暴力检测任务的多模态视频数据集 XD-Violence。
| **类别** | **类名** | **样本数** |
| -------- | -------- | ---------- |
| 0 | 正常 | 50 |
| 1 | 虐待 | 25 |
| 2 | 车祸 | 25 |
| 3 | 爆炸 | 25 |
| 4 | 斗殴 | 25 |
| 5 | 暴乱 | 25 |
| 6 | 枪击 | 25 |
## 使用建议
本数据集适用于探索性研究、内容审核模型原型、跨模态敏感内容检测与真实/伪造视频二分类等任务。由于文件夹布局并非标准 `datasets` 配置,建议先构建清单文件,将相对路径、模态、一级类别和细分类显式记录,再划分训练、验证和测试集。
使用前请注意:
- 不要将文件名、文本中的身份证号、电话、地址、银行卡号或其他疑似个人信息视为可公开再分发的数据;应先完成脱敏、来源核验和合规审查。
- 数据没有提供来源、授权、采集时间、标注流程、质量指标或推荐的数据划分说明。因此不应将其直接视为基准测试集,也不宜据此报告可泛化的性能结论。
- 训练、展示和人工审核应采用内容分级、最小权限、访问日志和适当的心理健康保护措施。
## 引用
如果您发现 Blockchain-Sensitive-Detect-Data 对您的研究或应用有帮助,请使用以下 BibTeX 格式进行引用:
```bibtex
@misc{anyang_song_2026,
author = { Anyang Song and Changkai Ji and Hao Fang and Yufeng Liu and Yang Li and Junjie He and Ying Cheng and Ruiwei Zhao and Yuejie Zhang and Rui Feng },
title = { Blockchain-Sensitive-Detect },
year = 2026,
url = { https://huggingface.co/datasets/anyangsong/Blockchain-Sensitive-Detect-Data },
doi = { 10.57967/hf/10293 },
publisher = { Hugging Face }
}
```