台灣公司的社會與治理指標與其勞動與職場安全健康法規遵循狀況之相關性分析

Abstract

本研究旨在探討台灣上市櫃公司永續報告書中，有關工作者安全與健康福祉的揭露內容，並檢視其與企業實際勞動法遵紀錄之關聯性。研究以2023年度企業永續報告書為主要語料來源，運用自然語言處理技術進行文本斷詞、詞頻分析與TF-IDF，描繪各產業揭露內容之語詞特徵與差異。本研究同時建構語意比對模型，量化報告書內容與《職場永續健康與安全SDGs揭露實務建議指南－2024修訂版》之語意相似度，進一步與勞動部公布之重大職業災害與違法紀錄相關性分析，檢視企業在揭露與實踐層面之間是否存在可觀察的對應關係。 研究結果顯示，企業揭露內容多傾向標準化語言，不同產業間雖具若干詞彙差異，但整體語用趨勢收斂。法遵紀錄則呈現明顯產業落差，營造業與重工業為違法與職災風險集中區，但其永續揭露比例偏低。語意相似度與違法紀錄之相關性不顯著，顯示揭露內容未必能真實反映企業職場安全健康的實務表現，惟語意得分高者在制度接軌與政策對應面仍具參考價值。 本研究提出四項政策建議：一、強化實質性揭露要求；二、建立語意監測工具；三、納入外包與供應鏈風險資訊；四、擴大高風險產業之揭露義務。研究結果除補充永續揭露與勞動實務間關係之實證證據，亦對職場安全與健康治理政策之資訊透明化、責任可歸屬與制度優化提供參考依據。

This study investigates how Taiwanese listed companies disclose information on occupational safety and health (OSH) in their 2023 sustainability reports, and examines whether such disclosures align with actual compliance records. Using natural language processing techniques—such as word segmentation, term frequency analysis, and TF-IDF weighting—this research identifies linguistic patterns across industries. A semantic similarity model was developed to assess alignment between report content and the 2024 Guideline on Sustainable Workplace Safety and Health SDG Disclosure. These scores were then cross-referenced with official records of major occupational accidents and labor law violations. Results show that most disclosures rely on standardized language, with only minor industry-level variation. While construction and heavy industries accounted for the highest compliance risks, they were underrepresented in sustainability reporting. No significant correlation was found between semantic similarity and compliance records, suggesting that disclosure content may not reflect actual OSH performance. However, high similarity scores may still indicate better alignment with institutional norms. The study proposes four recommendations: enhancing substantive disclosure, building semantic monitoring tools, including supply chain risks, and expanding reporting obligations for high-risk industries.