Document Key Value Extraction via Joint-Candidate Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for automatically extracting key values from documents rely on finding corresponding labels and understanding their location, which can lead to inaccuracies due to variations in label names and document templates, especially when documents from different sources or languages are involved.
Innovation Solution
The method involves identifying candidate values, generating joint-candidate sets, and using a trained machine learning mechanism to score these sets based on vertical and horizontal features, allowing for accurate key-value extraction without relying on label correlations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If key/value pattern matching techniques are used to extract information from documents, then the extraction process can be automated, but the accuracy of extraction deteriorates due to variations in label names and document templates across different sources and languages
Solution Approach 1:
The patent creates a universal extraction system that handles multiple document types, languages, and label variations through a single platform. The system uses machine learning models trained on diverse document data to recognize and extract key values regardless of the specific document template or language, making the extraction process universally applicable across different employers and document formats while maintaining high accuracy
Solution Approach 2:
The system dynamically adjusts extraction parameters based on the specific document being processed. Instead of using fixed pattern matching rules, the machine learning model adapts to different label names, formats, and document structures by changing its recognition parameters in real-time, allowing accurate extraction even when label names or template layouts vary between documents
2Ease of manufacture
If label-based pattern matching is used to find key values, then the extraction process is straightforward, but reliability deteriorates when document templates change or labels are positioned differently
Solution Approach 1:
The patent transforms the static label-based matching approach into a dynamic machine learning-based system. The extraction process adapts to different document templates and label positions by using trained models that can recognize key values regardless of their location or the specific template used, maintaining reliability even when document formats change
Solution Approach 2:
The system uses trained machine learning models that have learned from numerous example documents to create a generalized understanding of key value patterns. Instead of relying on exact copies of specific label patterns, the model creates abstract representations of key value relationships that can be applied across varying document templates, ensuring consistent extraction reliability
3Measurement precision
If manual entry of key values is used, then extraction accuracy can be high, but productivity deteriorates due to the tedious and error-prone nature of manual transcription
Solution Approach 1:
The patent implements an automated extraction system that performs the key value extraction task itself without requiring manual intervention. The machine learning model automatically identifies and extracts key values from document images, converting them into structured data that can be directly used by downstream applications, eliminating the need for manual transcription while maintaining high accuracy and significantly improving productivity
Data Source
AI summary
Techniques are described for extracting key values from a document without having to rely on finding corresponding labels for the target keys within the extracted text of the document. Further the techniques do not rely on knowledge of the correlation between (a) the location of labels within a document, and (b) the location of the key values that correspond to the labels. Key values are extracted from a document by, identifying candidate values within the document, establishing “joint-candidate” sets from those candidate values, and using a trained machine learning mechanism to score each joint-candidate set of values. The highest scoring joint-candidate set is deemed to reflect the correct mapping of candidate values to target keys for the document.