Document Data Extraction Using Reference Representation Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data extraction technologies face challenges in accurately extracting relevant data from electronic documents due to variations in form and content, leading to high inaccuracy and resource-intensive deep neural networks that require large amounts of training data and computing resources.
Innovation Solution
A data extraction system generates reference document representations by modifying unique text portions of reference documents, allowing for comparison with target documents to identify similar representations and apply associated extraction models for accurate data extraction, reducing the need for extensive training data and computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If complicated machine learning classification models such as deep neural networks are used to identify and extract data from electronic documents, then the system can handle various document formats, but the accuracy is high and false positive rates occur and large amounts of training data and computing resources are required
Solution Approach 1:
The patent segments the document processing task into multiple stages: document classification to identify document type, template selection to choose appropriate extraction pattern, and data extraction to pull specific information. This segmentation allows the system to handle diverse document formats through classification while maintaining high accuracy through specialized templates for each document type, avoiding the need for single complex models that produce false positives
Solution Approach 2:
The patent changes the approach from using complex model parameters (deep neural networks) to using structured parameters (document templates with defined fields and patterns). By parameterizing the extraction process around document-specific templates rather than generic deep learning models, the system achieves high accuracy across different formats without requiring large training datasets
2Adaptability or versatility
If complicated machine learning classification models are used to extract data from electronic documents, then various document formats can be processed, but large amounts of training data are required which is difficult to obtain
Solution Approach 1:
The patent performs preliminary classification to identify the document type before extraction, allowing the system to select the appropriate pre-defined template for that document type. This preliminary action enables the system to handle various document formats using specialized templates created in advance, eliminating the need for large amounts of training data that would otherwise be required to teach a deep learning model each format
Solution Approach 2:
The patent uses document templates that can be copied and adapted for different document types. Once a template is created for a particular document format, it can be replicated and modified for similar formats, allowing the system to handle diverse documents without requiring extensive training data for each variation
3Adaptability or versatility
If complicated machine learning classification models are used to extract data from electronic documents, then various document formats can be handled, but large amounts of computing resources are required to train or create the models
Solution Approach 1:
The patent segments the processing architecture into a lightweight classification component that identifies document type and a template-based extraction component that handles the actual data extraction. This segmentation eliminates the need for computationally intensive deep learning models, reducing energy consumption while maintaining the ability to handle various document formats through simple classification and pattern matching
Solution Approach 2:
The patent replaces expensive, resource-intensive deep learning models with inexpensive, simple template-matching algorithms. These templates are lightweight computational objects that can be stored and executed with minimal computing resources, providing the same document format handling capability without the high energy cost of training and running complex neural networks
Data Source
AI summary
Methods, systems, and computer storage media are provided for data extraction. A target document representation may be generated based on modified text of a target electronic document. A measure of similarity may be determined between the target document representation and a reference document representation, which may be based on modified text of a reference electronic document. Based on the measure of similarity, the reference document representation may be selected. An extraction model associated with the selected reference document representation can then be used to extract data from the target document.