Not sure if you've already found this arxiv preprint of an IEEE-published paper on LLM knowledge graph population yet: arXiv:2306.08302 It's about the field of using models to generate knowledge graphs from literature, and using these knowledge graphs as a knowledgebase to feed to a model trouble is that the models are constructed to replicate patterns, and it can easily make stuff that resembles meaningful literature but doesn't contain correct information. I would make sure to go over the output with a finetooth comb to make sure the values actually match the source material, outside of the obvious outliers where it's creating entirely irrelevant structured data - like what component metals go into a material, crystal structure, or similar.
Keep going bravo
Not sure if you've already found this arxiv preprint of an IEEE-published paper on LLM knowledge graph population yet: arXiv:2306.08302
It's about the field of using models to generate knowledge graphs from literature, and using these knowledge graphs as a knowledgebase to feed to a model
trouble is that the models are constructed to replicate patterns, and it can easily make stuff that resembles meaningful literature but doesn't contain correct information. I would make sure to go over the output with a finetooth comb to make sure the values actually match the source material, outside of the obvious outliers where it's creating entirely irrelevant structured data - like what component metals go into a material, crystal structure, or similar.
Such a great video. Thanks a lot :) i am working on this so this helps a lot
@@Pingu_astrocat21 glad to help!