Dataset Card for Juick.com Posts

Dataset Summary

This dataset contains 1,793,936 messages collected from Juick.com, a Russian microblogging service. Juick.com is a social platform where users share short messages, images, and interact through likes, comments, and recommendations. The dataset includes message text, user metadata, timestamps, tags, reactions, and attached images.

Languages

The dataset is primarily monolingual:

Russian (ru): The majority of messages and user content are in Russian, with some English content.

Dataset Structure

Data Files

The dataset consists of:

Metadata files: 4 compressed JSONL files (*.jsonl.zst) with 448,484 lines each, totaling 1,793,936 entries
Image files stored across multiple ZIP archives named attachments_00.zip, attachments_01.zip, etc., with approximately 5,000 images per archive

Data Fields

Each entry in the JSONL files contains the following fields:

id: Unique identifier for the post
response: Array containing post data with the following structure:
- user: User information object
  - avatar: URL to user's avatar image
  - uid: User ID
  - uname: Username
- updated: ISO timestamp of last update
- to: Target user information (for replies)
  - uname: Target username
- reactions: Array of reaction objects (optional)
  - id: Reaction type ID
  - description: Reaction description (e.g., "like")
  - count: Number of this reaction type
- recommendations: Array of users who recommended the post (optional)
  - avatar: Recommender's avatar URL
  - uid: Recommender's user ID
  - uname: Recommender's username
- likes: Total number of likes
- mid: Post ID
- body: Post text content
- tags: Array of strings representing hashtags (optional)
- timestamp: Post creation timestamp
- updated_at: Last update timestamp

Data Splits

The dataset contains a single split:

Split	Description	Number of Examples
`train`	Microblog posts and metadata	1,793,936

Total dataset size: 1,793,936 entries

image image