README.md

---
language:
  - zh
license: mit
pretty_name: RedNote Covert Advertisement Detection Dataset
tags:
  - covert advertisement detection
  - social-media
  - image-text
  - multimodal
  - RedNote
  - Xiaohongshu
datasets:
  - Jingyi77/CHASM-Covert_Advertisement_on_RedNote
dataset_info:
  - config_name: default
    features:
      - name: id
        dtype: string
      - name: title
        dtype: string
      - name: description
        dtype: string
      - name: date
        dtype: string
      - name: comments
        sequence:
          dtype: string
      - name: images
        sequence:
          dtype: string
      - name: image_count
        dtype: int32
      - name: label
        dtype: int8
      - name: split
        dtype: string
configs:
  - config_name: Example
    data_files:
      - split: Example
        path: example.parquet
      - split: Train_1
        path: train_part_1.parquet
      - split: Train_2
        path: train_part_2.parquet
      - split: Train_3
        path: train_part_3.parquet
      - split: Train_4
        path: train_part_4.parquet
      - split: Test
        path: test.parquet
      - split: Validation
        path: validation.parquet
size_categories:
  - 10<n<100
---

<!-- @format -->

# RedNote Covert Advertisement Detection Dataset

This dataset contains posts from the RedNote platform for covert advertisement detection tasks.

## Dataset Overview

| Split      | Posts    | Ad Posts | Non-Ad Posts | Total Images |
| ---------- | -------- | -------- | ------------ | ------------ |
| Train      | 3493     | 426      | 3067         | 18543        |
| Validation | 499      | 57       | 442          | 2678         |
| Test       | 1000     | 130      | 870          | 5103         |
| **Total**  | **4992** | **613**  | **4379**     | **26324**    |

> Note: The viewer shows a **small example subset** of the data (60 samples) for demonstration purposes. The complete dataset is available via WebDataset format in the repository.

## Field Descriptions

The example parquet file contains the following fields:

- `id`: Unique identifier for each post
- `title`: Post title
- `description`: Post description content
- `date`: Publication date (format: MM-DD)
- `comments`: List of comments
- `images`: List of base64-encoded images
- `image_count`: Number of images
- `label`: Label (0=non-advertisement, 1=advertisement)
- `split`: Data split (train/validation/test)

## Dataset Features

- **Multimodal Data**: Each post contains both text (title, description, comments) and images
- **Real-world Data**: Collected from actual social media posts on the RedNote platform
- **Multiple Images**: Each post may contain multiple images (average of 5.27 images per post)

## Data Format

The complete dataset is stored in WebDataset format, with each sample containing:

1. One or more image files (.jpg format)
2. A JSON metadata file with the following fields:
   - `id`: Sample ID
   - `title`: Title
   - `description`: Description
   - `date`: Date
   - `comments`: List of comments
   - `label`: Label (0: non-advertisement, 1: advertisement)

## Loading the Dataset

```python
from datasets import load_dataset

# Load example dataset
dataset = load_dataset("Jingyi77/CHASM-Covert_Advertisement_on_RedNote")

# Access a sample
example = dataset[0]
metadata = {
    "id": example["id"],
    "title": example["title"],
    "description": example["description"],
    "label": example["label"]
}
images = example["images"]  # List of images
```

## Citation

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

```
@dataset{CHASM,
  author = {Jingyi Zheng, Tianyi Hu, Yule Liu, Zhen Sun, Zongmin Zhang, Wenhan Dong, Zifan Peng, Xinlei He},
  title = {CHASM: Unveiling Covert Advertisements on Chinese Social Media},
  year = {2025},
  publisher = {Hugging Face},
  journal = {Hugging Face Hub},
  howpublished = {\url{https://huggingface.co/datasets/Jingyi77/CHASM-Covert_Advertisement_on_RedNote}}
}
```