You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

妃愛 & 新卡池角色 对话数据集

简介

包含多个动漫角色对话的单轮 QA 数据集,适用于角色扮演微调。

数据源

  • 妃愛路线_对话记录.md - 妃愛角色对话
  • 今天开新卡池了_对话记录.md - 多个角色对话

角色列表

角色 性格标签 训练集 测试集 合计
妃愛 genki, moe 2347 265 2612
華乃 moe 577 65 642
詩桜 genki 418 47 465
あすみ tsundere 322 37 359
dandere 297 34 331
天梨 genki 224 25 249
広夢 moe 72 9 81
莉々子 dandere 26 3 29

合计: 训练集 4283 条,测试集 479 条

数据格式

{
  "character": "妃愛",
  "trait": "genki",
  "question": "妃爱,早上好啊……",
  "dialogue": "呣哈……哥哥,早上好啊。今天早上也变大了呢"
}
  • character: 角色名
  • trait: 性格标签(genki/moe/dandere/tsundere/yandere-adj/kawaider/other)
  • question: 对话上下文(上一位发言者的台词)
  • dialogue: 角色回复

性格标签说明

标签 含义 对应角色
genki 元气 妃愛, 詩桜, 天梨
moe 萌系 妃愛, 華乃, 広夢
dandere 内向 里, 莉々子
tsundere 傲娇 あすみ
yandere-adj 病娇倾向 妃愛(部分)
kawaider 冷淡 妃愛(部分)

使用方法

import json

with open("combined_train.json", "r", encoding="utf-8") as f:
    train_data = json.load(f)

for record in train_data:
    print(f"{record['character']}: {record['dialogue']}")

文件结构

himeai_dataset/
├── combined_train.json      # 合并训练集 (4283)
├── combined_test.json       # 合并测试集 (479)
├── himeai_single_turn_train.json  # 原始妃愛训练集 (1802)
├── himeai_single_turn_test.json   # 原始妃愛测试集 (201)
├── 妃愛_train.json / 妃愛_test.json     # 妃愛单角色
├── 華乃_train.json / 華乃_test.json     # 華乃单角色
├── 詩桜_train.json / 詩桜_test.json     # 詩桜单角色
├── あすみ_train.json / あすみ_test.json  # あすみ单角色
├── 里_train.json / 里_test.json         # 里单角色
├── 天梨_train.json / 天梨_test.json     # 天梨单角色
├── 広夢_train.json / 広夢_test.json     # 広夢单角色
├── 莉々子_train.json / 莉々子_test.json  # 莉々子单角色
└── README.md

注意事项

  • 部分妃愛对话包含成人内容,使用时请注意
  • 测试集比例为 10%
Downloads last month
11