---
language:
- en
license: mit
task_categories:
- text-classification
- text-generation
task_ids:
- multi-class-classification
- language-modeling
size_categories:
- 10K<n<100K
tags:
- 4chan
- internet-culture
- text-data
pretty_name: 4chan /pol/ Dataset
pretty: true
dataset_info:
  features:
  - name: id
    dtype: int64
  - name: thread_id
    dtype: int64
  - name: board
    dtype: string
  - name: timestamp
    dtype: timestamp[ns]
  - name: title
    dtype: string
  - name: text
    dtype: string
  - name: text_length
    dtype: int64
  - name: filename
    dtype: string
  - name: file_ext
    dtype: string
  - name: file_size
    dtype: int64
  - name: image_width
    dtype: int64
  - name: image_height
    dtype: int64
  - name: is_op
    dtype: bool
  - name: mentions
    sequence: string
  - name: mention_count
    dtype: int64
  - name: replies
    dtype: int64
  - name: images
    dtype: int64
  - name: unique_ips
    dtype: int64
  - name: content_hash
    dtype: string
  - name: archived
    dtype: bool
  - name: semantic_url
    dtype: string
  - name: hour_of_day
    dtype: int32
  - name: day_of_week
    dtype: string
  - name: is_weekend
    dtype: bool
  - name: post_count
    dtype: int64
  - name: total_images
    dtype: int64
  - name: avg_text_length
    dtype: float64
  - name: std_text_length
    dtype: float64
  - name: total_mentions
    dtype: int64
  splits:
  - name: train
    num_bytes: 122600567
    num_examples: 317418
  download_size: 56680481
  dataset_size: 122600567
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---



# 4chan /pol/ dataset

This dataset contains data from 12000+ threads from 4chan boards, collected and processed for research purposes. The data includes both active and archived threads, with extensive metadata and derived features for studying online discourse and community dynamics.  
I preserved thread structure, temporal information, and user interaction patterns while maintaining anonymity and excluding sensitive content.

## Dataset Details

### Dataset Description

- **Curated by:** [vmfunc](https://x.com/vmfunc)
- **Language(s):** English
- **License:** MIT
- **Size:** [Dataset size]
- **Number of Records:** 
  - Posts: 12435

### Dataset Sources and Creation

The dataset was collected using a custom scraper that interacts with 4chan's official API.  
Active threads are collected in real-time. Only publicly available data is collected and no PII is available. This data collection adheres to 4chan's robots.txt and API guidelines.  
The collection process follows these steps:  

### Collection Process
1. **Active Threads**: First, the scraper fetches the catalog of currently active threads from the board's catalog.json endpoint.

2. **Archive Integration**: The scraper then queries the board's archive.json endpoint to obtain a list of archived thread IDs, ensuring comprehensive coverage of both current and historical content.

3. **Thread Processing**: For each thread (both active and archived):
   - All posts within the thread are collected
   - HTML entities are decoded and tags are stripped while preserving line breaks
   - Mentions (>>post_numbers) are extracted and tracked
   - Post metadata (timestamps, file information, etc.) is preserved
   - Thread-level metrics are calculated (reply count, unique posters, etc.)
   - Temporal features are derived
   - Everything is processed in parallel (ThreadPoolExecutor)
   - MD5 hashing is used to identify and remove duplicate posts
   - All collected data is validated against a predefined schema

## Uses

### Direct Use

The dataset is suitable for:
- Studying online discourse patterns and community dynamics
- Analyzing temporal patterns in online discussions
- Research on thread structure and user interaction patterns
- Natural language processing tasks on informal internet communication
- Content analysis and topic modeling
- Network analysis of post references and replies

### Out-of-Scope Use

This dataset should not be used for:
- Identifying or tracking individual users
- Generating harmful or malicious content
- Training models for harassment or abuse
- Analyzing sensitive personal information

## Dataset Structure

### Data Fields

#### Post-level Features
- `id`: Unique post identifier (int64)
- `thread_id`: Thread identifier (int64)
- `board`: Board identifier (string)
- `timestamp`: ISO format timestamp (timestamp[ns])
- `title`: Thread or post title (string)
- `text`: Clean post text content (string)
- `text_length`: Length of the post text (int64)
- `filename`: Original filename (string)
- `file_ext`: File extension (string)
- `file_size`: Size of attached file in bytes (int64)
- `image_width`: Width of attached image (int64)
- `image_height`: Height of attached image (int64)
- `is_op`: Boolean indicating if post is the original post (bool)
- `mentions`: List of post references (list)
- `mention_count`: Number of mentions in the post (int64)
- `replies`: Number of replies (int64)
- `images`: Number of images (int64)
- `unique_ips`: Number of unique IPs in thread (int64)
- `content_hash`: MD5 hash of post content (string)
- `archived`: Boolean indicating if thread is archived (bool)
- `semantic_url`: Thread's semantic URL (string)

#### Thread-level Features
- `post_count`: Total posts in thread (int64)
- `total_images`: Total images in thread (int64)
- `avg_text_length`: Average text length in thread (float64)
- `std_text_length`: Standard deviation of text length (float64)
- `total_mentions`: Total mentions in thread (int64)

#### Temporal Features
- `hour_of_day`: Hour when post was made (int64)
- `day_of_week`: Day of the week (string)
- `is_weekend`: Boolean indicating weekend posts (bool)

## Personal and Sensitive Information

- All data is from public boards only
- No IP addresses or unique identifiers included
- Content hashing used for deduplication
- No personal information preserved

## Bias, Risks, and Limitations

### Technical Limitations
- Incomplete thread relationships due to archival
- Missing posts due to deletion
- Temporal gaps in archived content
- File contents not included

### Biases
- Selection bias from board choice
- Survivorship bias from archived content
- Temporal bias from collection period
- Community-specific language patterns

### Risks
- Potential for offensive content
- Risk of harmful pattern learning
- Bias in language models trained on the data

## Dataset Card Authors

[vmfunc](https://x.com/vmfunc)

## Dataset Card Contact

[vmfunc](https://x.com/vmfunc)