This article discusses the combination of Knowledge Graphs and Large Language Models (LLMs) to enable users to ask their own questions and receive comprehensive answers. While Knowledge Graphs are effective at representing domain data and delivering answers through expert-formulated queries, LLMs allow any user to ask questions but lack domain-specific information. The article demonstrates how a knowledge graph can prompt or fine-tune an LLM using an RDF knowledge graph of a process plant. It also explores the challenges of querying the RDF graph using SPARQL and the potential of LLMs to provide accurate answers. The article concludes by discussing scaling options such as pre-indexing and fine-tuning with the RDF graph.
| Signal | Change | 10y horizon | Driving force |
|---|---|---|---|
| Knowledge Graphs + Large Language Models | Ability to ask questions to LLM | LLMs can provide comprehensive answers | Integration of domain knowledge |
| Users need to ask questions to databases | Enhanced user ability to ask questions | Users can ask their own questions | Improved user experience |
| Knowledge Graphs represent domain data | Improved data representation | More accurate and comprehensive data representation | Better decision-making and analysis |
| Large Language Models provide comprehensive answers | Increased answer accuracy | More accurate and detailed answers | Enhanced information retrieval |
| RDF knowledge graphs can prompt or fine-tune LLMs | Improved LLM performance | More efficient and accurate LLM responses | Better integration of graph and LLM |
| Pre-indexing RDF graph for LLM | Scalability of LLM with large graphs | LLMs can handle larger and more complex graphs | Efficient retrieval of relevant information |
| Fine-tuning LLM with RDF graph | Enhanced LLM performance and accuracy | LLMs can provide more accurate domain-specific answers | Better integration of domain knowledge |
| Knowledge Graphs as precursor to LLM training data | Improved LLM training data | Better training and performance of LLMs | Enhanced performance of LLMs |
| Integration of Knowledge Graphs and LLMs | Increased value of domain knowledge | Enhanced utilization of domain knowledge | Improved decision-making and insights |
| Improvements in LLM technology | Advancements in LLM capabilities | More advanced and powerful LLM models | Technological innovation |