Full text loading...
We propose a human-in-the-loop AI approach to translate WITSML mnemonics into D-WIS ontology terms for improved interoperability. We explore the possibility of leveraging the log probability features of large language models (LLMs) to estimate model confidence in its selections. Our results show that we can achieve nearly 100% accuracy by reducing the number of user choices from hundreds to just ten. However, during our study, we encountered a major issue regarding the completeness of the D-WIS model and the poor quality of many mnemonic descriptions.