Using Machine Learning on Unstructured Text for Knowledge Extraction: What if you could read all of the reports?

Jean-Philippe Paiement, P.Geo., M.Sc., Director, Global Consulting - Mira Geoscience

Jean-Philippe Paiement is the Director of Global Consulting at Mira Geoscience Ltd. He brings 15 years of mineral exploration experience to the table, including expertise in geostatistics applied to structural, geological, and geochemical modelling and interpretation; specializing in non-linear interpolation and simulation. Jean-Philippe has developed multiple workflow and novel approaches to reduce interpretational risks of geological data. Jean-Philippe has a wide range of experience in mineral resource estimation for precious metals, base metals and industrial minerals across diverse geological environments around the world. In 2016, Jean-Philippe has pioneered the application of Machine Learning to the mineral exploration industry in winning the Integra GoldRush challenge by application of machine learning to mineral deposit targeting. He is skilled in the application of machine learning to overcome geological and geophysical challenges; by combining geological knowledge and both supervised learning and deep learning. Before joining Mira Geoscience, he obtained an MSc from Laval University. Jean-Philippe is based in Quebec-City.

Talk Abstract:

Geoscientific papers and reports have been compiled and published on numerous online platforms in the last 20 years, which make for a substantial knowledge resource. All these written text document represent the common knowledge of mineralized systems since the begging of modern geosciences. However, the sheer amount of information makes it prohibiting to anyone looking at compiling and summarizing this information.

Recent advances in Natural Language Processing (NLP), a subfield of artificial intelligence and linguistics as shown promising results in speech recognition. Using speech tagging algorithms, sentences from text documents can easily be divided between root verbs, subjects, and objects. This information can then be summarized into knowledge graphs consisting of a network of nodes, representing subjects/objects and edges representing relationships between nodes. 

In this talk, we will investigate how we could use the combination of NLP and graph theory to conduct knowledge extraction from district scale papers to property scale reports. The simplification and easy visualization of this substantial amount of public knowledge might represent the next leap in exploration and targeting model construction. This novel approach enables the users to access key information from all predeceasing work of a given research/exploration subject.

Live Online Broadcast:

Please register/log-in for this online broadcast using the link above. TGDG members will also be able to view this presentation at a later date by visiting our member's only Talk Recordings page.