How to integrate an existing structured source or an existing graph?

[SotirisKot]

Hi everyone,

My use case is related in integrating existing knowledge graphs into GraphRAG. More concretely, I have an ontology describing fields of science.

Examples:
{
"level_1": "engineering and technology",
"level_2": "electrical engineering, electronic engineering, information engineering",
"level_3": "artificial intelligence",
"level_4": "natural language processing",
"level_5": "neural machine translation"
},
{
"level_1": "medical and health sciences",
"level_2": "clinical medicine",
"level_3": "oncology & carcinogenesis",
"level_4": "oncology",
"level_5": "granular cell tumor"
}

As you can see the ontology is already hierarchical and each level is connected to its lower level. Now for the lowest levels of the ontology, I can also provide related scientific publications, descriptions of each field of science, journals/conferences or authors that publish to these fields of science and as it is evident this can turn to a knowledge graph, where you can pose questions.

The above information is not in unstructured format where GraphRAG can perform its indexing. Is it possible for me to force the LLM that performs the indexing to include all the already existing structured information, without losing any fields or extra metadata?

I was thinking of creating text chunks that specifically describe the ontology. For example:
Text Chunk 1: "

• engineering and technology is a parent to electrical engineering, electronic engineering, information engineering
• electrical engineering, electronic engineering, information engineering is a parent to artificial intelligence
• ...
• more relations
• ...
• Description of the field of natural language processing: description text
• Related scientific publications to natural language processing: publication list
• w/e other info I have until I reach the token threshold

"
Text Chunk 2: "text for the second chunk and so on"

Would that work? Do you have any other suggestions? Maybe providing this information to the prompt?

Thanks a lot for your time!