Datasets:
Dataset Card for tw-highschool
本資料集收錄臺灣高中四大主要學科(生物、化學、數學、物理)之繁體中文知識文本,作為訓練具備臺灣高中學科知識能力之繁中模型的補充語料。
Dataset Details
Dataset Description
資料以「學科」為 config 拆分:
biology:生物chemistry:化學math:數學physics:物理
每個 config 內含對應學科之知識性段落(涵蓋觀念解說、公式推導、範例題、術語對照等),以繁體中文書面語為主。可作為高中學科 chatbot、教學助理之預訓練/微調補強。
- Curated by: Huang Liang Hsun
- Language(s) (NLP): Traditional Chinese
- License: cc-by-nc-sa-4.0
Dataset Sources
- Repository: lianghsun/tw-highschool
- Chat version: lianghsun/tw-highschool-chat
- Paper: TBA
Uses
Direct Use
- 補強模型在高中四大學科上的知識覆蓋。
- 衍生下游解題、教學 chatbot 之 SFT/RAG 來源。
Out-of-Scope Use
- 不適用作為現行課綱的事實依據;課綱與教材會逐年更新。
- 不應作為考試標準答案來源;本資料以知識覆蓋為主,並非校勘版本。
Dataset Structure
每個 config 內 train split 結構大致如下:
{
"text": "..."
}
(實際欄位以資料集載入後 features 為準)。
Dataset Creation
Curation Rationale
高中學科知識在通用網頁語料中常以零碎方式出現,不利於模型學習有系統的觀念。本資料集以「學科 → 章節 → 段落」的方式整理,提升模型在高中學科上的回答品質。
Source Data
Data Collection and Processing
- 蒐集臺灣高中各學科之知識文本(教材、講義、政府公開教材等)。
- 以學科分目錄存放,移除版面雜訊與圖片參考。
- 為每個 config 輸出獨立 jsonl。
Who are the source data producers?
原始教材由各教科書出版社/教師/政府公開資源提供。
Annotations
Annotation process
無人工標註。
Who are the annotators?
無人工標註者。
Personal and Sensitive Information
不含個人識別資料。
Bias, Risks, and Limitations
- 學科內容反映特定編寫年代之觀點,與最新教材/研究可能略有差異。
- 各 config 規模未必平均,下游使用時可能需要再做樣本平衡。
Recommendations
建議與 tw-highschool-chat(chat 版本)一同使用,將「知識」與「對話」兩層一起訓練。
Citation
@misc{tw_highschool,
title = {tw-highschool: Taiwan High School Knowledge Corpus},
author = {Huang, Liang Hsun},
year = {2025},
howpublished = {\url{https://huggingface.co/datasets/lianghsun/tw-highschool}}
}
Dataset Card Authors
Dataset Card Contact
- Downloads last month
- 8