Predicting hydrocarbon plays from text using machine learning and natural language processing. I recently tested the OpportunityFinder Algorithm on a selection of public domain geoscience literature. Only literature published between 1990 to 2010 was used, some time before a major gas discovery was made in the area. The hypothesis was whether the algorithm could surface the existence of a ‘play type’ and supporting evidence way in advance to its impending discovery.
The algorithm surfaced the play type of [RESERVOIR]-[TRAP]-[SEAL] of Miocene Shallow Water Limestones in Atoll-like reef structures capped by thick salt in the area. Similar in nature to what was subsequently found through exploration. Evidence of gas was picked up through seafloor pockmarks, present where the salt was absent or via faults. There were no pockmarks at the vicinity of the discovery, showing evidence for potentially a good seal and gas trapped below the salt. This perhaps hints as to what may be achieved through these types of text mining techniques.
No claims are made that algorithms can provide an ‘x’ marks the spot. AI is generally unimaginative and lacks the retroductive reasoning of a geoscientist. However, what algorithms can do, is ‘read’ more reports & papers than a person can feasibly read in a lifetime; joining the dots to surface subtle potentially interesting patterns to spark ideas. These suggestions may point the geoscientist towards a line of thinking they may not have had otherwise. We may be just scratching at the surface of what is possible.