--- license: apache-2.0 language: - zh task_categories: - text-generation tags: - legal - Taiwan - ROC - zh-tw - law-article - pretrain - cpt pretty_name: 中華民國台灣法規全文後處理資料集 size_categories: - 10K tw-processed-law-ctx 是一個中華民國(台灣)法規全文之後處理資料集,合計 11,462 筆。相較於 [tw-processed-law-article](https://huggingface.co/datasets/lianghsun/tw-processed-law-article)(以單一條文為單位),本資料集將「同一法規之所有條文」合併為單一 `text`,提供整部法規之完整上下文,適合作為繁中法律 LLM 之持續預訓練(CPT)素材。 ## Dataset Details ### Dataset Description 以「條文為單位」之語料雖便於精確查詢,但模型在訓練時難以學到整部法規之章節結構與條文之間之關聯。本資料集將每一部法規之所有條文依序合併為單一長文本,保留: * 法規名稱(含章節標題); * 所有條文之條號與本文; * 該法規最近修正日期與 API 更新日期; * 若法規已廢止,額外提供 `abandon_note` 欄位。 每筆亦同時附帶 `level`(法規位階:憲法 / 法律 / 命令 / 廢止)作為篩選依據。 - **Curated by:** [Liang Hsun Huang](https://www.linkedin.com/in/lianghsunhuang/?locale=en_US) - **Language(s) (NLP):** Traditional Chinese - **License:** Apache 2.0 - **Source Authority:** [全國法規資料庫](https://law.moj.gov.tw/) ### Dataset Sources - **Repository:** [lianghsun/tw-processed-law-ctx](https://huggingface.co/datasets/lianghsun/tw-processed-law-ctx/) - **Related:** [lianghsun/tw-processed-law-article](https://huggingface.co/datasets/lianghsun/tw-processed-law-article) - **Source Authority:** [全國法規資料庫](https://law.moj.gov.tw/) ## Uses ### Direct Use 本資料集主要設計用於: * 繁體中文法律 LLM 之持續預訓練(CPT); * 訓練模型理解整部法規之章節結構與條文之間之關聯; * 作為法律語料 pipeline 中之「長上下文」組成部分; * 與條文單位之語料(如 `tw-processed-law-article`、`tw-structured-law-article`)互補。 ### Out-of-Scope Use 本資料集不適用於下列用途: * 作為正式法律諮詢之依據; * 作為條文單位之精準查詢語料(合併後難以定位單一條號); * 作為非中華民國法律體系之訓練資料。 ## Dataset Structure ```json { "text": "中華民國憲法 第 一 章 總綱\n中華民國憲法 第 1 條 中華民國基於三民主義...\n中華民國憲法 第 2 條 中華民國之主權屬於國民全體。\n...", "name": "中華民國憲法", "level": "憲法", "abandon_note": "", "modified_date": "19470101", "api_updated_date": "2024/1/12 上午 12:00:00" } ``` | 欄位 | 型別 | 說明 | |---|---|---| | `text` | string | 整部法規之所有條文合併後之連續文本 | | `name` | string | 法規名稱 | | `level` | string | 法規位階(憲法 / 法律 / 命令 / 廢止) | | `abandon_note` | string | 廢止備註(若法規已廢止) | | `modified_date` | string | 最近修正日期 | | `api_updated_date` | string | API 更新時間 | | 統計項目 | 數值 | |---|---| | 總筆數 | 11,462 | | 切分 | 無(使用者自行切分) | ## Dataset Creation ### Curation Rationale `tw-processed-law-article` 以條文為單位切分,雖便於查詢,但在 CPT 階段無法提供長上下文。本資料集將同一法規之條文合併,使模型能學到整部法規之章節組織與條文間之語境。 ### Source Data #### Data Collection and Processing - **來源**:全國法規資料庫之法規 API; - **處理**:將 `tw-processed-law-article` 之條文依法規名稱分組並合併,保留章節標題與條號。 #### Who are the source data producers? 原始法規由中華民國立法院、行政院及各主管機關制定並公告,並由法務部於全國法規資料庫維護。 ### Annotations #### Annotation process 本資料集不包含額外人工標註。 #### Who are the annotators? 不適用。 #### Personal and Sensitive Information 本資料集為公開之法規條文,不包含個人敏感資訊。 ## Bias, Risks, and Limitations * **時效性**:法規會持續修正,本資料集反映特定時點之版本快照; * **長文本偏態**:整部法規合併後文本長度差異極大,自普通條文(幾百字)到複雜法律(數萬字)皆有; * **廢止法規亦包含**:需依 `level` 與 `abandon_note` 進行篩選。 ### Recommendations 建議使用者: * CPT 訓練時依 `level` 過濾(例如僅使用「法律」與「命令」,排除已廢止); * 與條文單位語料(`tw-processed-law-article`、`tw-structured-law-article`)搭配使用,同時獲得長上下文與精確條文; * 使用時定期與全國法規資料庫比對以更新訓練資料。 ## Citation ```bibtex @misc{tw-processed-law-ctx, title = {tw-processed-law-ctx: ROC Full-Law Context Corpus}, author = {Liang Hsun Huang}, year = {2024}, howpublished = {\url{https://huggingface.co/datasets/lianghsun/tw-processed-law-ctx}}, note = {11,462 ROC laws with all articles merged into a single text per law, for CPT training.} } ``` ## Dataset Card Authors [Liang Hsun Huang](https://www.linkedin.com/in/lianghsunhuang/?locale=en_US) ## Dataset Card Contact [Liang Hsun Huang](https://www.linkedin.com/in/lianghsunhuang/?locale=en_US)