Language-Model Data Modeling for Form Structure Extraction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Rendering forms into computer-usable data models is challenging due to the difficulty in capturing the structure and transforming raw text into a specific data structure, as existing methods fail to extract and summarize form information effectively.

Innovation Solution

A data model generator leveraging a language model to process raw data structures by extracting text, determining text types, generating prompts for enhanced information, and defining a data model with a combination of text and enhanced information, using components like data extraction controllers, language models, and data model generators.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If screen scraping application is used to extract text from form, then text extraction is achieved, but form structure and relationships are not captured

Engineering Contradiction:

Improveform structure informationVSAvoidtext extraction simplicity

Core Design Contradiction:

Loss of informationVSEase of operation

Solution Approach 1:

The system segments the form processing into multiple stages: initial text extraction, text type classification, enhanced information generation, and data model construction. Each stage handles specific aspects of form information, preserving both raw text and structural context through systematic division of the processing workflow.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary language model that acts as a bridge between raw extracted text and structured data models. This intermediary component generates enhanced information including form summaries, field definitions, and relationship descriptions, thereby preserving structural information that would otherwise be lost in direct extraction.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Ease of manufacture

If raw text extraction is performed, then text data is obtained, but the text is not in a format readily transformable into data structure

Engineering Contradiction:

Improvedata structure transformationVSAvoidformatting information

Core Design Contradiction:

Ease of manufactureVSLoss of information

Solution Approach 1:

The system performs preliminary actions by classifying extracted text into different types (narrative, tabular, list, key-value) and generating enhanced information before final data model construction. This preliminary processing prepares the text in advance for easier transformation into structured data formats by identifying appropriate structural patterns early in the workflow.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent applies parameter changes by transforming text characteristics through the language model, which modifies the textual representation into enhanced information with different structural parameters. The language model changes the state of text from unstructured raw data to enhanced information with defined properties such as summaries, field definitions, and relationship descriptions.

Inventive Principle:

Principle #35Parameter changes

3Ease of operation

If no form summary is available, then processing is simpler, but working with raw text becomes more difficult

Engineering Contradiction:

Improveraw text processingVSAvoidprocessing system complexity

Core Design Contradiction:

Ease of operationVSDevice complexity

Solution Approach 1:

The system implements self-service by using the extracted text itself as input to generate the form summary through the language model. The same extracted text that would otherwise remain unprocessed is automatically transformed into enhanced information including summaries, field definitions, and relationship descriptions, eliminating the need for separate manual summarization processes.

Inventive Principle:

Principle #25Self-service

4Measurement precision

If multiple text types are processed with different prompts, then enhanced information accuracy is improved, but processing time increases

Engineering Contradiction:

Improveenhanced information accuracyVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system applies partial action by selectively processing different text types with appropriate prompts rather than applying uniform processing to all text. Text classification identifies which portions of the extracted text require enhanced information generation, allowing the system to focus computational resources on specific text segments that benefit most from language model processing, thereby reducing overall processing time while maintaining accuracy.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS12572516B2Data model generator leveraging a language model

Publication Date: 2026.03.10 INTUIT INC

AI summary

One or more embodiments provide for a method. The method includes applying, to a raw data structure defining a form, a data extraction controller to extract text from the raw data structure. The method also includes determining a text type of the text. The method also includes generating, according to the text type, a prompt for a language model to request enhanced information. The enhanced information includes information implicit in the form. The method also includes applying, with the prompt, the language model to the text to generate the enhanced information. The method also includes applying a data model generator to define a data model with a combination of the text and the enhanced information. The data model is a computer-readable representation of the form.