Consensus Value Determination for Duplicate Document Fields
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document extraction systems face inaccuracies when dealing with documents containing multiple copies of the same information, as they rely on raw confidence scores from OCR services, leading to inconsistencies and user dissatisfaction.
Innovation Solution
The system employs a combination of rule-based analytical techniques and machine learning methods to determine consensus values for repeated fields, using modules like text extraction, confidence score analysis, featurization, summarization, answer-ranking, and named entity recognition to predict final values accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the extraction engine selects the key-value pair with the highest confidence score as the final extracted value, then the selection process is simple and fast, but the accuracy of value determination decreases when duplicate fields exist
Solution Approach 1:
The patent segments the selection process into multiple stages: initial extraction of all candidate key-value pairs, confidence score filtering to identify high-confidence candidates, and consensus determination through comparison of duplicate fields. This multi-stage segmentation allows the system to maintain speed while improving accuracy by systematically evaluating multiple candidates rather than relying on a single highest-score selection.
Solution Approach 2:
The patent implements feedback mechanisms where the system compares extracted values across duplicate fields and uses consensus determination to refine the final selected value. The confidence scores serve as feedback to identify which fields warrant closer examination, and the consensus process provides feedback to resolve inconsistencies, thereby improving overall accuracy while maintaining efficient processing.
2Loss of information
If the system extracts multiple key-value pairs from every region of a document with acceptable quality, then the completeness of data extraction increases, but the complexity of value determination increases
Solution Approach 1:
The patent applies preliminary action by performing confidence score analysis on all extracted key-value pairs before final selection. The system pre-identifies high-confidence candidates and pre-groups duplicate fields, which simplifies the subsequent consensus determination process. This preliminary organization of data reduces the complexity of the final selection stage while maintaining complete data extraction.
Solution Approach 2:
The patent changes parameters by introducing confidence score thresholds and consensus criteria to manage the complexity of value determination. By adjusting these parameters, the system can control the level of detail in the analysis while ensuring complete data extraction. The parameter-based approach allows flexible management of extraction completeness versus determination complexity.
3Productivity
If the extraction engine relies on raw confidence scores from OCR services, then the processing is straightforward and fast, but the representativeness of the confidence scores for extraction accuracy decreases
Solution Approach 1:
The patent introduces an intermediary consensus determination process between the raw OCR confidence scores and the final extracted value. This intermediary layer analyzes multiple candidate values, compares their confidence scores, and determines the most representative value through consensus. This mediator improves the representativeness of the final confidence assessment while maintaining processing efficiency by using systematic rather than ad-hoc evaluation methods.
Data Source
AI summary
Systems and methods are provided to determine consensus values for duplicate fields in a document or form.