Image Reading Ensemble Validation for Unstructured Entity Extraction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current image reading systems struggle with converting unstructured form-like documents into structured digital formats, requiring manual intervention and suffering from inaccuracies due to unstructured data, spelling errors, and challenges in entity extraction and grouping, particularly in financial documents like loan notices.

Innovation Solution

An image reading system utilizing a machine learning kernel to generate bounding boxes and search paths for textual blocks, combined with a descriptive linguistics engine to identify target textual block pairs, enabling automated entity extraction and validation without human intervention.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If traditional OCR mechanisms are used to extract data from unstructured form-like documents, then the extraction process can be automated, but the accuracy and reliability of extracted entities deteriorate due to unstructured data, spelling errors, and ambiguities

Engineering Contradiction:

Improveautomation of entity extractionVSAvoidaccuracy of extracted entities

Core Design Contradiction:

Extent of automationVSReliability

Solution Approach 1:

The patent introduces an intermediary validation process that cross-checks extracted entities against multiple data sources including system of record databases, external APIs, and contextual information from the document itself. This intermediary layer filters and verifies extracted entities before final output, resolving the contradiction between automation and reliability by adding a verification buffer between extraction and delivery.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system implements feedback loops where extraction results are validated against known data patterns, system of record information, and contextual constraints. When inconsistencies are detected, the system automatically adjusts extraction parameters or flags entities for manual review, creating a continuous improvement cycle that maintains high reliability while preserving automation.

Inventive Principle:

Principle #23Feedback

2Reliability

If manual entity extraction is performed to ensure accuracy, then the reliability of extracted entities improves, but the productivity and speed of processing deteriorate due to time-consuming manual intervention

Engineering Contradiction:

Improveaccuracy of extracted entitiesVSAvoidprocessing speed

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The system applies partial automation by using OCR and extraction algorithms for initial entity identification, then applying validation rules and cross-referencing only to the extracted entities rather than processing entire documents manually. This selective approach maintains high speed while ensuring accuracy through targeted verification of critical data points.

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The extraction process is segmented into multiple independent stages: initial extraction, validation against system of record, contextual verification, and final output. Each segment can be processed with appropriate level of automation or manual review, allowing the system to optimize for both speed in automated segments and accuracy in critical verification segments.

Inventive Principle:

Principle #1Segmentation

3Device complexity

If generic OCR processing is used on unstructured documents, then the device complexity remains low, but the measurement precision of extracted entities deteriorates due to inability to understand unstructured data context

Engineering Contradiction:

Improvesimplicity of processing systemVSAvoidaccuracy of entity identification

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The system employs a universal extraction framework that can handle multiple document types and unstructured formats through a single processing pipeline. The same core extraction engine adapts to different document structures by applying configurable validation rules and contextual analysis, maintaining simplicity while improving precision through multi-functional processing capabilities.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20260080706A1Image reading systems, methods and storage medium for performing entity extraction, grouping and validation

Publication Date: 2026.03.19 US BANK NATIONAL ASSOCIATION

AI summary

Entity extraction is performed using a voting ensemble approach to enhance the accuracy and effectiveness of entity extraction processes. Voting ensemble weights are stored. In turn, a prediction value for a first entity is determined from multiple models that are specifically trained for entity extraction. A plurality of first entity prediction values is obtained. The model with the highest prediction value among the plurality of first entity prediction values is then selected.