A Generative AI-Powered ESG Survey Engine: A Semantic Search, Retrieval-Augmented Generation and Content Analysis Approach

In this project, we have gained a clear understanding of the significance of Environmental, Social, and Governance (ESG) for enterprises. Simultaneously, we have witnessed the exceptional performance of Generative AI (GenAI) in various tasks. Therefore, we are committed to further exploring the application of GenAI across diverse ESG-related tasks within enterprises. In this phase, building upon the progress achieved in the initial two phases, which encompassed the development of esgBERT for machine reading ESG articles and the creation of a ChatGPT-based tool for generating ESG news summaries, we have established three primary research objectives. Firstly, our objective is to assess a Semantic Search Algorithm (SSA) that can assist proponents of ESG initiatives in tasks such as identifying similar ESG cases and planning ESG indicators. Additionally, we aim to evaluate the performance of Large Language Models (LLMs) when employing Retrieval-Augmented Generation (RAG) through Content Analysis (CA) for drafting press releases. Secondly, we seek to improve the accuracy of generated ESG summaries and ensure their compliance with standards. To accomplish this, we plan to evaluate content generated with open-source Large Language Models (LLMs) using content analysis methods as a potential alternative to ChatGPT. Lastly, our aim is to package all the outputs and provide operational manuals to facilitate a seamless transition for the Applied Materials Technology (AMT) team as they take on these responsibilities.