iammytoo/japanese-humor-evaluation

Japanese Multimodal Humor Evaluation Dataset

This dataset combines two Japanese humor datasets for evaluating the funniness of responses to prompts (odai).

Dataset Description

This dataset merges:

• —bokete dataset: Image prompts with text responses
• —keitai dataset: Text prompts with text responses

All scores are normalized to a 0-4 scale for consistency.

Dataset Structure

Data Fields

• —odai_id: Unique identifier for the prompt
• —odai_type: Type of prompt ('image' or 'text')
• —odai: The prompt content (image reference for bokete, text for keitai)
• —response: The humorous response text
• —score: Normalized humor score (0-4 scale)
• —original_score: Original score from the source dataset
• —original_dataset: Source dataset ('bokete' or 'keitai')
• —split: Original data split
• —user_name: Responder name (keitai only)
• —award: Award received (keitai only)

Data Splits

The dataset is split into train/validation/test sets with an 80/10/10 ratio.

Usage

from datasets import load_dataset

dataset = load_dataset("iammytoo/japanese-humor-evaluation")

# Access different splits
train_data = dataset['train']
val_data = dataset['validation']
test_data = dataset['test']

# Example item
print(train_data[0])

Source Data

• —bokete dataset: https://huggingface.co/datasets/YANS-official/ogiri-bokete
• —keitai dataset: https://huggingface.co/datasets/YANS-official/ogiri-keitai

Citation

If you use this dataset, please cite the original datasets:

@misc{japanese-humor-evaluation,
  title={Japanese Multimodal Humor Evaluation Dataset},
  author={iammytoo},
  year={2024},
  publisher={HuggingFace}
}

License

This dataset is released under the Apache 2.0 license, following the original datasets' licensing.