README.md

metadata

annotations_creators:
  - no-annotation
language_creators:
  - found
language:
  - en
  - ru
license: unknown
multimodal:
  - text
  - tabular
pretty_name: Shikimori Dataset
size_categories:
  - n_10K
source_datasets:
  - original
task_categories:
  - tabular-regression
task_ids:
  - tabular-multi-class-classification

Shikimori Dataset

Anime database from Shikimori with user ratings, suitable for recommender systems and ML research.

Dataset Structure

The dataset contains 3 separate data files:

anime.parquet / anime.jsonl (9,950 entries)

Anime titles with metadata.

Field	Type	Description
id	int	Shikimori anime ID
name	string	English name
russian	string	Russian name
description	string	Synopsis (BBCode)
score	float	Average community score
rating	string	Age rating (g, pg_13, r, etc.)
episodes	int	Total episodes
episodes_aired	int	Aired episodes (for ongoing)
duration	int	Episode duration in minutes
genres	dict	Genre IDs and names
season	string	Season (e.g. fall_2023)
aired_on	string	Premiere date
released_on	string	Release date
status	string	released/ongoing/anons
studios	dict	Studio info
is_censored	bool	Censored flag

users_rates.parquet / users_rates.jsonl (67,071 entries)

User anime viewing history and ratings.

Field	Type	Description
id	int	Rate ID
user_id	int	Shikimori user ID
anime_id	int	Anime ID
score	int	User's score (0 = unset)
episodes	int	Episodes watched
rewatches	int	Rewatch count
created_at	string	First rate timestamp
updated_at	string	Last update timestamp

genres.parquet / genres.jsonl (80 entries)

Genre reference table.

Field	Type	Description
id	int	Genre ID
name	string	English name
russian	string	Russian name

Use Cases

• Recommender Systems (Collaborative Filtering, Matrix Factorization)
• Anime Classification by genre/description
• Score Prediction models
• User Behavior Analysis
• LLM fine-tuning for anime recommendations

Data Source

Original scraped from Shikimori API. See: https://shikimori.one/api/doc

Notes

• Anime descriptions contain BBCode formatting ([character=id]name[/character])
• Scores are community averages (float, 1-10 scale)
• Users are anonymized by ID (not by name)
• Dataset snapshot date: April 2026