Knowledge Augmented Intelligence Using Large Language Models for Advanced Data Analytics

Abstract Large language model (LLM) holds the potential to facilitate advanced data analysis within the realm of Oil and Gas (O&G) operations by delivering cutting-edge performance in Natural Language Processing (NLP) tasks such as data mining in text. In this paper, we propose a novel data-driven Artificial Intelligence Augmented Intelligence strategy with LLMs to facilitate the exploration of topics for causal machine learning to understand choices under considerations. State-of-the-art LLMs are finetuned, evaluated, benchmarked, and put into practical use to understand and monitor evolving narratives and perspectives over time for trends and quantitative analytics. Experiments show that model performance is optimized through supervised fine-tuning based on domain specific data. Methods for evaluating the fine-tuning quality and assessing semantic relevance are introduced. Fine-tuned LLMs categorize topics in the target documents, while machine learning technics normalize guideline documents to establish a baseline for benchmarking. LLMs are efficient in processing vast volumes of industry-specific data and extracting valuable insights. In the O&G industry, LLMs have the potential to improve human intelligence via Knowledge Augment Intelligence through qualitative-to-quantitative analytics.