Financial Event Extraction via Segmented NLP Pipeline

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Automatically discerning relational and event information about entities in unstructured electronic documents, such as financial documents, is difficult due to language variations and the unstructured nature of the content, even with state-of-the-art computing equipment.

Innovation Solution

A system for named-entity tagging and event extraction that uses a combination of tokenization, zoning, and statistical tagging, along with authority files and machine-learning classifiers, to identify and classify entities and relationships in text, constraining the search space with high-precision classifiers and integrating multiple sources of knowledge for enhanced accuracy and efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If state-of-the-art computing equipment is used for automatic information extraction, then processing speed is improved, but accuracy and reliability remain insufficient due to language variations and unstructured content

Engineering Contradiction:
Improveprocessing speedVSAvoidextraction accuracy
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The system segments the information extraction process into distinct stages: tokenization, zoning, statistical tagging, and relationship extraction. Each stage processes specific aspects of the text independently, allowing optimization of each component while maintaining overall accuracy and speed.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs preliminary tokenization and zoning before the main extraction process. By pre-processing the text into tokens and identifying zones of interest, the system prepares the data structure in advance, improving both processing speed and subsequent extraction accuracy.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If manual analysis of unstructured documents is used, then extraction accuracy is maintained, but processing time and labor requirements increase significantly

Engineering Contradiction:
Improveextraction accuracyVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system enables self-service processing by automatically performing tokenization, zoning, and entity extraction without requiring manual intervention. The statistical tagger and relationship extractor operate autonomously on the processed text, dramatically reducing processing time while maintaining accuracy through sophisticated algorithms.

Inventive Principle:
Principle #25Self-service

3Reliability

If complex processing algorithms are applied to handle language variations, then extraction reliability is improved, but device complexity and processing overhead increase

Engineering Contradiction:
Improvehandling language variationsVSAvoidprocessing complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system breaks down the complex language processing into separate modules: tokenization handles word segmentation, zoning identifies relevant text regions, statistical tagging handles language variations, and relationship extraction identifies entities and connections. This modular approach manages complexity while improving reliability.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The statistical tagger dynamically adjusts its processing parameters based on the input text characteristics and language detected. By adapting parameters such as tokenization granularity and tagging sensitivity to the specific document, the system handles language variations effectively without requiring overly complex fixed algorithms.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP2257896B1Financial event and relationship extraction
Publication Date: 2021.07.14 THOMSON REUTERS ENTERPRISE CENTRE GMBH
  • EP2257896B1 patent drawingFigure 1
  • EP2257896B1 patent drawingFigure 2
  • EP2257896B1 patent drawingFigure 3

AI summary

For automated text processing, the inventors devised, among other things, an exemplary system (100) that automatically extracts financial events from various unstructured text based sources, such as press releases and news articles. Extracted events, such as mergers & acquisitions, earnings guidance reports, and actual earnings announcements, are represented as structured data records which can be linked, searched, and displayed and used as a basis for controlling accessing to the source documents and other related financial documents for named entities.