ihounie/when2call_imbalanced_request
收藏Hugging Face2026-03-17 更新2026-03-29 收录
下载链接:
https://hf-mirror.com/datasets/ihounie/when2call_imbalanced_request
下载链接
链接失效反馈官方服务:
资源简介:
---
pretty_name: when2call_imbalanced_request
configs:
- config_name: train_pref
data_files:
- split: train
path: train-*
license: mit
language:
- en
tags:
- when2call
- preference-dataset
- class-imbalance
- synthetic-sampling
size_categories:
- 1K<n<10K
---
# when2call_imbalanced_request
This dataset is derived from `nvidia/When2Call` (`train_pref`, `train` split) by downsampling one chosen-response category to ~50% while keeping all other rows.
## Source
- Dataset: `nvidia/When2Call`
- Config: `train_pref`
- Split: `train`
- Source rows: 9000
## Classification Rules (on `chosen_response`)
Categories are assigned in this precedence order:
1. `toolcall` if text contains `<TOOLCALL>` (case-insensitive)
2. `request` if text contains `?`
3. `request` if text contains one of:
- `To proceed,`
- `Please provide`
- `Please specify`
(case-insensitive)
4. `refusal` if text contains one of:
- `apologies`
- `apologize`
- `sorry`
- `I'm unable` (including escaped/quoted variants)
- `I'm afraid`
(case-insensitive)
5. otherwise `unk`
## Sampling Procedure
- Target minority class: `request`
- Keep ratio for target class: 50% (floor when odd)
- Random seed: 44
- Other classes: all rows kept
## Class Counts (chosen_response)
### Before sampling
- refusal: 2999
- toolcall: 3000
- request: 3001
- unk: 0
### After sampling
- refusal: 2999
- toolcall: 3000
- request: 1500
- unk: 0
## Rows
- Final rows: 7499
## Notes
- The schema/columns match the source `train_pref` split format.
- This repo contains only the `train_pref`/`train` data.
pretty_name: when2call_imbalanced_request(不平衡呼叫请求数据集)
configs:
- config_name: train_pref
data_files:
- split: train
path: train-*
license: MIT
language:
- en
tags:
- when2call
- preference-dataset(偏好数据集)
- class-imbalance(类别不平衡数据集)
- synthetic-sampling(合成采样数据集)
size_categories:
- 1000 < 样本量 < 10000
# when2call_imbalanced_request(不平衡呼叫请求数据集)
本数据集源自`nvidia/When2Call`的`train_pref`训练拆分版本,通过对某一选定的回复类别进行下采样至约50%的占比,其余类别样本全部保留。
## 数据集来源
- 原始数据集:`nvidia/When2Call`
- 原始配置:`train_pref`
- 原始拆分:训练集(train)
- 原始样本量:9000
## 选定回复(chosen_response)的分类规则
分类将按照以下优先级依次判定:
1. 若文本包含`<TOOLCALL>`(不区分大小写),则归类为`toolcall`(工具调用类)
2. 若文本包含问号`?`,则归类为`request`(请求类)
3. 若文本包含以下任一内容(不区分大小写),则归类为`request`(请求类):
- `To proceed,`
- `Please provide`
- `Please specify`
4. 若文本包含以下任一内容(不区分大小写),则归类为`refusal`(拒绝类):
- `apologies`
- `apologize`
- `sorry`
- `I'm unable`(包括转义或带引号的变体)
- `I'm afraid`
5. 其余情况归类为`unk`(未知类)
## 采样流程
- 目标少数类:`request`(请求类)
- 目标类保留比例:50%(样本量为奇数时向下取整)
- 随机种子:44
- 其余类别:全部保留原样本
## 选定回复的类别分布
### 采样前
- 拒绝类(refusal):2999条
- 工具调用类(toolcall):3000条
- 请求类(request):3001条
- 未知类(unk):0条
### 采样后
- 拒绝类(refusal):2999条
- 工具调用类(toolcall):3000条
- 请求类(request):1500条
- 未知类(unk):0条
## 最终样本量
- 最终总样本量:7499条
## 说明
- 数据集的结构(Schema)与字段与原始`train_pref`拆分版本一致。
- 本仓库仅包含`train_pref`/`train`拆分的数据集。
提供机构:
ihounie


