The article compares the performance of large language models (LLMs) Darwin, LLaMA, and LLaMA2 in named entity recognition (NER), relation extraction (RE), and entity resolution (ER) within materials science. Darwin shows superior performance, achieving higher F1 scores in NER and RE compared to its counterparts. However, all models perform similarly in ER tasks, possibly due to limited contextual memory in LLMs. To address this issue, entity resolution techniques are applied to improve standardization tasks of the inference outputs, enhancing overall performance.
Darwin achieves markedly higher F1 scores in both NER and RE tasks than the LLaMA 7b and LLaMA2 7b models, indicating better performance in materials science.
Despite better performance in NER and RE, there was no significant difference for ER tasks, possibly due to LLMs' insufficient contextual memory capabilities.
#large-language-models #natural-language-processing #materials-science #entity-recognition #ai-performance-review
Collection
[
|
...
]