This paper presents the development of a recursive deep learning framework utilizing natural language processing and computer vision techniques for optimizing oil drilling operations. Following the CRISP-DM methodology, the objectives include extracting both structured and unstructured data from various oilfield drilling report documents for sentiment analysis, training models capable of autonomously understanding hidden information in text/images to determine performance accuracy, leveraging daily drilling report analysis to enhance model accuracy and mitigate non-productive time risks, thus improving drilling operation efficiency, and creating a user-friendly interface for seamless interaction with the model. The results and discussion encompass predicted values, comparison between actual and predicted values, as well as scattered and line plots visualizing the relationship between actual and predicted values.

Keywords: Recursive Deep Learning, Natural Language Processing, Computer Vision, Oil Drilling, Sentiment Analysis, CRISP-DM, Model Training, Efficiency Optimization