Convolutional Neural Network for Document Extraction Confidence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Evaluating the correctness of extracted document information is challenging, especially when different extraction models are used, as it is complex to determine if the extracted output from rule-based or machine learning models matches the original information, and this evaluation is necessary for ensuring correctness and understanding the accuracy of the extraction models.
Innovation Solution
A system that utilizes a convolutional neural network (CNN) to process extracted document information and the document itself, generating confidence and correctness scores to evaluate and correct the extracted information, and is agnostic to the extraction model used, employing techniques like denoising autoencoders for feature selection and error detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If rule-based or machine learning models are used to extract document information, then extraction capability is improved, but evaluating the correctness of extracted information becomes complex and challenging
Solution Approach 1:
A CNN-based evaluation model is introduced as an intermediary component that receives both the original document and extracted information as inputs. This mediator automatically assesses the correctness of extraction results by comparing them against the original document, thereby simplifying the evaluation process while maintaining high accuracy. The evaluation model acts as a bridge between extraction models and final output validation.
Solution Approach 2:
The system implements a feedback mechanism where the CNN evaluation model generates correctness scores and confidence values that are fed back to assess extraction quality. This feedback loop enables automatic validation of extracted information, allowing the system to identify and correct errors without manual intervention, thus reducing evaluation complexity while improving reliability.
2Measurement precision
If multiple extraction models are used to handle different document complexities, then extraction accuracy is improved, but determining correctness across different models becomes more complex
Solution Approach 1:
The CNN-based evaluation model is designed as a universal assessment tool that can evaluate extraction results from any extraction model (rule-based or machine learning) regardless of the document complexity. It processes both the original document and extracted information through a unified architecture, generating standardized correctness scores and confidence values that work across different model types and document classes.
3Reliability
If extracted information is evaluated against original document information, then correctness assurance is improved, but the evaluation process becomes complex
Solution Approach 1:
The manual or complex rule-based evaluation process is replaced with a CNN-based automated evaluation system. The neural network automatically compares extracted information against the original document, learning the correct relationships and patterns during training. This substitution transforms a complex mechanical evaluation process into an automated intelligent system that provides reliable correctness assessment with simplified operations.
Data Source
AI summary
Disclosed herein are system, method, and computer program product embodiments for correcting extracted document information based on generated confidence and correctness scores. In an embodiment, a document correcting system may receive a document and document information that represents information extracted from the document. The document correcting system may determine the correctness of the document information by processing the document to generate a character grid representing textual information and spatial arrangements for the text within the document. The document correcting system may apply a convolutional neural network on character grid and the document information. The convolutional neural network may output corrected document information, a correctness value indicating the possible errors in the document information, and a confidence value indicating a likelihood of the possible errors.