The ESG Knowledge Hub
06/05/2025
This section introduces the ESG Knowledge Hub, a conversational AI tool that leverages GraphRAG and a domain-specific knowledge graph to deliver accurate, explainable, and context-aware answers in the ESG domain. It outlines how the tool enhances traditional generative AI by integrating query expansion, user profiles, and relevant recommendations.
This section introduces the ESG Knowledge Hub, summarizing how the tool uses a domain-specific knowledge graph and GraphRAG to improve traditional generative AI.
Background
Traditional conversational AI platforms have increased in popularity over recent years. They have been deemed as useful due to their ability to answer questions in a natural, convincing and concise manner. However, the results retrieved from these generative AI models are often inaccurate and cannot be traced back to their source. This is particularly dangerous in workspaces or other cases where the accuracy, transparency and explainability of information needs to be guaranteed. For this reason, RAG (Retrieval Augmented Generation) has been developed to tackle these issues.
GraphRAG combines retrieval-augmented generation with graph-based representations—such as knowledge graphs or other structured data—to produce concise, understandable as well as accurate and traceable information. This is critical when it comes to the complex landscape of ESG reporting. Knowledge-hub.eco is a conversational AI demo application based on the ESG Core Knowledge Model which is currently the most comprehensive domain-specific knowledge model available for the ESG domain. The tool uses concepts extracted by the knowledge model to expand the user query, thereby minimizing hallucination in the output generated by the LLM. It then uses GraphRAG for its Main takeaways section.
How it works
The ESG Knowledge Hub improves traditional generative AI in the following ways:
First, the user query is expanded with concepts from the ESG Core Knowledge Model. The user can choose concepts from the provided auto-complete feature, which suggests available narrower concepts to specify the question if desired. This greatly reduces hallucination due to the relevant context that is added to the query.
Further context is provided by the user's desired user profile, thereby taking user intent into account and increasing the relevance of the generated output.
The PoolParty Extractor extracts concepts from the generated summary. The most relevant concepts and the footprint from the selected user profile are then used for content and document recommendation.
The Main takeaways section addresses the user's initial question. It is generated via GraphRAG using the recommended documents with the highest relevance.
Guide Outline
The following pages will guide you through the the tool's UI as well as illustrate how to interact with its generated output.