基於ESG風險管理設計一套流程結合ChatGPT自動生成新聞摘要及演講稿;Design a pipeline for ESG risk management combined with ChatGPT to automatically generate news summaries and cue cards

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/93019

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/93019

Title:	基於ESG風險管理設計一套流程結合ChatGPT自動生成新聞摘要及演講稿;Design a pipeline for ESG risk management combined with ChatGPT to automatically generate news summaries and cue cards
Authors:	王瑋;Wang, Wei
Contributors:	資訊工程學系
Keywords:	ESG風險管理;新聞摘要;演講稿生成;ChatGPT;ESG risk management;News summarization;Cue card generation;ChatGPT
Date:	2023-07-05
Issue Date:	2024-09-19 16:38:44 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著環境、社會和治理(Environmental, Social and Government, ESG)因素對企業日益重要，永續經營不再只是口號，而是一個更實際的目標。越來越多投資者開始將ESG指標納入評估項目，關注企業遇到ESG風險及機遇時的表現，另外過去文獻也指出良好的ESG表現與企業的財務績效呈現正相關。本研究的每種方法都以ChatGPT模型當作最後步驟，用以產生結果。首先，採用輿情分析方法，透過ChatGPT模型找出ESG相關新聞中的核心話題，接著比較四種不同pipeline流程產生的新聞摘要，包括Zero-shot GPT模型、兩輪式Zero-shot GPT模型、Extractive-Abstractive模型及Abstractive-Abstractive模型。同時，還比較了利用ChatGPT模型得到的客製化prompt搭配演講稿模板和直接使用一般prompt生成演講稿的效果，最終選擇能生成清晰敘述且高連貫性的兩輪式Zero-shot GPT模型作為新聞摘要的模型，並且選擇能穩定生成指定格式的客製化prompt搭配演講稿模板作為生成演講稿的指令。基於上述研究結果，本研究提出兩項自動化工具：新聞摘要產生器及演講稿產生器。這兩項工具被整合到演講稿生成系統中，使用者能透過新聞摘要產生器快速了解當週的ESG流行詞及相關議題的綜合新聞摘要，節省從大量新聞中尋找重要議題的時間，接著，使用者可以將回應該議題的內容作為新聞稿文本，並提供演講者名稱，就能透過演講稿產生器生成一份演講稿。這將使企業能有即時回應的能力，幫助企業減少ESG風險並提高ESG機遇。;With the increasing importance of Environmental, Social, and Governance (ESG) factors in businesses, sustainable operations have become more than just a slogan; they are now a practical goal. More and more investors are incorporating ESG indicators into their evaluation criteria, focusing on how companies perform in the face of ESG risks and opportunities. Past literature has also shown a positive correlation between good ESG performance and financial performance. Each method in this study employs the ChatGPT model as the final step to generate results. Firstly, sentiment analysis is utilized to identify the main topic in ESG-related news using the ChatGPT model. Then, four different pipeline processes for generating news summaries are compared, including the Zero-shot GPT model, the Two-stage Zero-shot GPT model, the Extractive-Abstractive model, and the Abstractive-Abstractive model. Additionally, the effectiveness of using customized prompts combined with cue card templates and using general prompts directly to generate cue card is compared. Ultimately, the Two-stage Zero-shot GPT model is chosen for news summaries due to its ability to generate clear and coherent descriptions, and a customized prompt combined with a cue card template is selected as the instruction for generating cue cards. Based on the aforementioned research findings, this study proposes two automation tools: a news summary generator and a cue card generator. These tools are integrated into a cue card generation system, allowing users to quickly grasp the comprehensive news summary of ESG trends and relevant topic for the week through the news summary generator. This saves time spent on searching for important issues from a large volume of news. Subsequently, users can input their responses to the topic as press release and provide the speaker′s name to generate a cue card through the cue card generator. This empowers businesses with the ability to respond promptly, helping them mitigate ESG risks and capitalize on ESG opportunities.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	57	View/Open

社群 sharing

Loading...