應用文字探勘與機器學習於企業永續之分析

近年因為新冠肺炎疫情，國際掀起了一波重視ESG主題的熱潮，也有愈來愈多研究指出ESG因素能影響投資表現，並且考慮ESG因素有望幫投資人帶來更佳的風險調整後的報酬表現，身為ESG領頭的歐盟，更是在近年積極推動各種ESG的相關規範以及約束，尤其是對於金融業與重污染的產業特別關注。國際間也開始陸續出現共識去遵從幾個被廣泛認可的原則或規範，例如赤道原則（Equator Principles，縮寫EPs）、氣候相關財務揭露(Task Force on Climate-related Financial Disclosures，縮寫TCFD)、碳揭露專案(Carbon Disclosure Project，縮寫CDP)等等。隨著世界各地對ESG的重視，國際間也出現了許多ESG的評級機構，最為知名幾家有MSCI（Morgan Stanley Capital International）、FTSE Russell、Sustainalytics等等，可惜的是這些機構評級的評分標準不全然相同，有時做出來的評級結果差異頗大。目前，國際尚未發展出一套擁有共識、全世界遵守的ESG關鍵績效指標（KPI），對於此塊領域仍在摸索和調整中。現今，台灣企業也努力想要跟進國際趨勢，並對永續議題貢獻一份力，可惜因為起步較晚，多數台灣企業的ESG數據、評級資訊常有缺失及不夠全面等問題，也導致不好判別出哪些是趁機「漂綠」的黑心企業，讓真正想投資ESG的投資者無所適從。因此本研究想要參考國內外已有的論文基礎，從台灣企業資料較完整的社會責任報告書去提取出有用的ESG資訊，利用文字探勘先將企業社會責任報告書中的文字擷取，再使用機器學習的技術去將文字做分類，最終期許能有效地推論出此企業ESG成績的好壞，使投資者或利害關係人不用苦於無企業的ESG評級數據，能藉由此研究的成果去進行決策考量之判斷。

關鍵字

文字探勘；機器學習； ESG永續；社會責任報告書；數據分析

並列摘要

In recent years, due to COVID-19, there has been a wave of international attention to ESG topics, and more and more studies have pointed out that ESG factors can affect investment performance, and considering ESG factors is expected to help investors bring better risk-adjusted return performance. In recent years, the European Union, which is the leader of ESG, has actively promoted various ESG-related norms and constraints, especially in the financial industry and heavily polluting industries. Consensus has also emerged internationally to comply with several widely recognized principles or norms, such as the Equator Principles (EPs), Task Force on Climate-related Financial Disclosures (TCFD), carbon disclosures Project (Carbon Disclosure Project, abbreviated CDP) and so on. With the emphasis on ESG around the world, many ESG rating agencies have emerged internationally. The most well-known ones are MSCI (Morgan Stanley Capital International), FTSE Russell, Sustainalytics, etc. It is a pity that the rating standards of these agencies are not the same.Sometimes the ratings are quite different. At present, the world has not yet developed a set of ESG key performance indicators (KPIs) that have a consensus and are followed by the world, and are still being explored and adjusted. Therefore, this study intends to refer to the existing papers at domestic and abroad to extract useful ESG information from the corporate sustainability report with relatively complete information of Taiwanese companies. Using machine learning technology to classify text, it hoped that it can effectively infer the ESG performance of the company, so that investors or stakeholders do not have to suffer from the ESG rating data of no company, and can use the results of this research to make judgments for decision-making considerations.