Improved Information Retrieval From Well Related Documents Using Supervised Learning

Abstract

We introduce a system for rapid retrieval of relevant well related information from a corpus of over 20 million documents. This allows for exploration workers to retrieve important business data more quickly. Tracking down all of the information required to make complex business decisions is a time consuming and error prone process. This poses a direct risk of expensive miscalculations and missed opportunities. A first version of this system is currently undergoing tests with select users. As the work here represents the first version of the system, it is expected that improvements will be made. This is a system that can be used at enterprise scale to enable searches to more easily yield usable information to workers. This system uses a supervised learning model to identify well related documents from several categories. Examples of these categories include (but are not limited to) formation evaluation and well completion reports. A machine learning model was trained to classify documents according to input from a well document expert. This input came in the form of a set of labeled documents compiled by said expert. This model was then applied to over 20 million documents that are deemed relevant to the exploration process. The inferred classifications for each document were stored in a search engine in order to facilitate retrieval of documents by each of the labels from above. The benefits of this system are twofold. First, it reduces the number of documents that come back for a given search of a large corpus of documents. Second, it allows users without technical experience in well-related work to more easily find documents.