Open wbcbugfree opened 5 months ago
We managed to figure out how to use domain-specific ontologies/vocabs when manually annotating RDF triples from text (see #8), but now the problem becomes how to get LLMs to do the same. There are two solution paths here:
We temporarily use post-processing to ensure that the RDF triples generated by LLMs do not contain any customized and non-existent ontologies.
Introduce domain-specific ontologies/vocabularies in place of generic/customized ones for concepts from soil health knowledge graph. Here are some candidates: