Datasets:
Tasks:
Text Generation
Modalities:
Text
Formats:
json
Languages:
Japanese
Size:
10K - 100K
Tags:
roleplay
License:
metadata
configs:
- config_name: default
data_files:
- split: filtered
path: Japanese-Roleplay-Dialogues-Filtered.jsonl
- split: original
path: Japanese-Roleplay-Dialogues-Original.jsonl
license: apache-2.0
task_categories:
- text-generation
language:
- ja
tags:
- roleplay
size_categories:
- 10K<n<100K
Japanese-Roleplay-Dialogues
This is a dialogue corpus collected from Japanese role-playing forum (commonly known as "なりきりチャット(narikiri chat)"). Each record corresponds to a single thread.
For the original
version, no filtering has been applied.
For the filtered
version, the following filtering and cleaning conditions have been applied:
- If the number of unique
poster
in theposts
of each record is 1 or less, delete the entire record. - If the length of the
posts
is 10 or less, delete the entire record. - Normalize poster names by removing all whitespace characters (both full-width and half-width spaces) and anonymizing them.
- Retain only records where the top two posters account for 90% or more of the total posts.
- Remove posts from posters other than the top two posters.
- Among the top two posters, neither should account for more than 60% of their combined posts. (i.e., each of the top two posters should account for at least 40% of their combined posts).
- Remove initial consecutive posts from the same poster until a different poster appears.
- Remove final consecutive posts from the same poster until a different poster appears.
- Merge consecutive posts from the same poster into a single post, separated by line breaks.
Not all dialogues are purely role-playing. Some records include initial discussions about the settings, or they may continue from other threads.
This dataset is intended for use in machine learning applications only.