# US 2022/0036008 A1 · Method and System for Generating Geological Lithostratigraphic Analogues using Theory-Guided Machine Learning from Unstructured Text

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2022/0036008 A1
- Document ID: US20220036008A1
- Status: Published application
- Kind: A1
- Application number: 16/944,425
- Filed: 2020-07-31
- Published: 2022-02-03
- Earliest claimed: 2020-07-31
- Assignees: GeoScienceWorld
- Inventors: Paul Hugh Cleverley
- Web page: https://patentgrep.com/patent/US20220036008A1

## CPC

- G06F 40/295: Named entity recognition
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06N 5/022: Knowledge engineering; Knowledge acquisition
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/00: Machine learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

The invention is a data processing method and system for assisting geoscientists in identifying geological Lithostratigraphic analogues from unstructured text using theory-guided machine learning. The data processing system makes the necessary calculations to detect geological Lithostratigraphic entities and appropriate entity relations, accounting for space (local geography) and time (local succession) when using word associations to determine similarity. The system computes the reliability of that score, incorporating user feedback into the learned model. In particular, the data processing system operates on any digital unstructured text derived from academic literature, company reports, web pages and other sources. Similarity between the associations of words to geological Lithostratigraphic entities can be used to suggest global analogues to communities such as petroleum geoscientists in order to reduce petroleum exploration risk and improve field development decisions.

## Claims

### Claim 1

A computer implemented process for suggesting space-time aware geological Lithostratigraphic analogues from text using theory-guided machine learning, the system comprising:
receiving into memory:
sentences of text, geographical lexicons, negation/negative terms, geological time lexicons and Natural Language Processing (NLP) rules,
processing the data in the computer memory with a processor to detect geological Lithostratigraphic entity names,
processing the data in the computer memory with a processor to compute similarity between Lithostratigraphic entities through word associations; and
outputting the Lithostratigraphic entity-entity similarity (through cosine similarity or other method), by:
applying a filter around the detected geological Lithostratigraphic entities, the filter having a dynamic width defined by a window length comprising the first mention of a Lithostratigraphic entity and a subsequent mention of a different Lithostratigraphic entity, word associations amalgamated where subsumed entity relations are inferred, removing geographical terms, the filter output providing an input to a learned statistical model weighting (from high to low) nouns, adjectives and verbs, from which Lithostratigraphic entity-entity similarity can be determined.

### Claim 2 (depends on 1)

The computer-implemented process of claim 1 further comprising:
Computing an entropy reliability function for each entity using the number of mentions of a Lithostratigraphic entity multiplied by the number of documents containing its said mentions multiplied by the number of unique authors/sources for documents containing its said mentions.

### Claim 3 (depends on 2)

The computer-implemented process of claim 2 further comprising:
receiving into memory:
human feedback from the presentation of Lithostratigraphic entity similarity score ranking
processing the data in computer memory using a computer processor to incorporate the user feedback into the Lithostratigraphic entity similarity calculation, by:
altering the learned statistical model for the Lithostratigraphic entities, through such techniques as (but not limited to) reduction or increase of the original input text and/or overall pairwise similarity score.