Document Key-Value Pair Extraction via ML Bounding Box Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for extracting data from documents are labor-intensive, prone to human error, and require significant manual effort, as they often rely on manual data entry or user-configured extraction rules, which are not adaptable to varying document layouts or structures.
Innovation Solution
The use of machine learning models to determine bounding boxes for key-value pairs in documents, generating mappings between these boxes based on their locations, allowing for automatic extraction and linking of data elements without user input, enabling efficient and accurate data extraction across different document structures and contents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual data entry is used to extract data from documents, then data can be entered into a database, but the process is tedious, slow, difficult to scale, subject to human error, and resource-intensive
Solution Approach 1:
The system uses machine learning models to automatically extract and link key-value pairs from documents without requiring manual data entry. The models self-adjust to different document layouts through training, enabling the system to serve itself in adapting to various document formats while maintaining high accuracy and speed.
Solution Approach 2:
The patent replaces the mechanical process of manual data entry with an automated machine learning-based system. The machine learning models detect and extract data elements from documents, substituting human manual operations with automated computational processes that are faster and more accurate.
2Adaptability or versatility
If user-configured extraction rules are used to convert text to structured data, then semantic context can be provided, but the rules cannot be used with documents that adhere to other layouts
Solution Approach 1:
The machine learning models are dynamic and can adapt to different document layouts through training. Instead of static extraction rules that work only for specific layouts, the models learn to identify data elements across various document structures, making the system adaptable to new layouts without requiring manual rule reconfiguration.
Solution Approach 2:
The machine learning-based extraction system serves multiple document layouts and formats universally. A single trained model can handle various document types (invoices, receipts, forms, etc.) with different structures, eliminating the need for separate extraction rules for each layout type.
3Measurement precision
If machine learning models are trained to predict bounding boxes based on content and ordering, then the models may perform well with similar documents, but fail to recognize data elements in documents with dissimilar content or structure
Solution Approach 1:
The system performs preliminary training with diverse document samples before deployment. By exposing the machine learning models to a wide variety of document layouts and structures during the training phase, the models learn to generalize their data extraction capabilities to handle dissimilar documents effectively in production.
Solution Approach 2:
The machine learning models adjust their internal parameters through training on diverse datasets. This parameter optimization enables the models to capture various document structure patterns, improving their ability to identify data elements across different document types while maintaining precision.
Data Source
AI summary
One embodiment of the present invention sets forth a technique for extracting data from a document. The technique includes determining, via execution of one or more machine learning models, a first set of bounding boxes for a first set of values associated with a first key within the document and a second set of bounding boxes for a second set of values associated with a second key within the document. The technique also includes generating a first set of mappings between a list of items in the document and the first set of bounding boxes and a second set of mappings between the first and second sets of bounding boxes based on locations of the bounding boxes. The technique further includes determining, for a given item, one or more associated bounding boxes in the first and second sets of bounding boxes based on the mappings.