Using the glossary to answer questions

[jsvgoncalves]

Goal: Explore how to combine the power of large language models (LLMs) with knowledge graphs derived from the LCA glossary to provide accurate and context-aware answers to LCA-related questions.

Key Questions:

• Knowledge Graph Representation: Is the current data model sufficiently fit for purpose or do need to move to e.g., using RDF, Neo4j?
• RAG Integration: What are the best practices for integrating LLMs with knowledge graphs (e.g., using frameworks like LangChain)?

Potential Outcomes:

• A prototype question-answering system that demonstrates the potential of RAG for LCA.
• A roadmap for further development and deployment of this system.

[tngTUDOR]

What is the RAG ?

[jsvgoncalves]

Here's a good source https://neo4j.com/developer-blog/fine-tuning-retrieval-augmented-generation/

[simontaurus]

We made some first experiences with reconstructing simple human language via SPARQL CONSTRUCT, serializing triples to short
"subject_label property_label object_label"
sentences, e. g. "9SMn28 is a subclass of Steel".
Latest LLMs like GPT4 are also able to generate SPARQL from text