Abstract
Drilling activities in the oil and gas industry have been reported overdecades for thousands of wells on a daily basis, yet the analysis of this textat large-scale for information retrieval, sequence mining, and pattern analysisis very challenging. Drilling reports contain interpretations written bydrillers from noting measurements in downhole sensors and surface equipment,and can be used for operation optimization and accident mitigation. In thisinitial work, a methodology is proposed for automatic classification ofsentences written in drilling reports into three relevant labels (EVENT,SYMPTOM and ACTION) for hundreds of wells in an actual field. Some of the mainchallenges in the text corpus were overcome, which include the high frequencyof technical symbols, mistyping/abbreviation of technical terms, and thepresence of incomplete sentences in the drilling reports. We obtainstate-of-the-art classification accuracy within this technical language andillustrate advanced queries enabled by the tool.