Document Entity Extraction With Spatial Grounding Validation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document entity extraction techniques are costly to instantiate and lack flexibility to parse new document types, often relying on hand-crafted parsers or specifically trained machine-learned models that are prone to errors and hallucinations.
Innovation Solution
A flexible document entity extraction system using a machine-learned model-agnostic approach that grounds extracted data to reference spatial locations within the document, employs consistency metrics and voting systems to validate outputs, and leverages existing model architectures for energy efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If hand-crafted parsers or specifically trained machine-learned models are used for document entity extraction, then extraction capability is achieved, but the system becomes costly to instantiate and lacks flexibility to parse new document types
Solution Approach 1:
The patent employs a universal machine-learned model architecture that can process multiple document types without requiring separate parsers or retraining. The model is designed to handle diverse document formats (invoices, receipts, forms, etc.) through a single unified system, eliminating the need for document-type-specific models and reducing instantiation costs while maintaining high adaptability to new document types.
2Reliability
If specifically trained machine-learned models are used, then extraction capability is achieved, but the models become prone to errors and hallucinations
Solution Approach 1:
The patent implements a feedback mechanism where the model's extraction outputs are validated against the original document content and spatial locations. The system checks whether extracted entities are grounded in the actual document, providing feedback to correct or reject hallucinated extractions. This validation loop significantly reduces errors and hallucinations while maintaining the model's adaptability to various document types.
Solution Approach 2:
The patent performs preliminary grounding of extracted entities to their spatial locations in the document before finalizing the extraction results. By pre-validating that each extracted entity corresponds to actual document content at the predicted location, the system prevents hallucinations from occurring in the first place, thereby improving reliability without sacrificing adaptability.
3Productivity
If traditional entity extraction methods are used, then data extraction is performed, but computational costs are high and energy efficiency is reduced
Solution Approach 1:
The patent extracts only the essential spatial location information from the machine-learned model outputs for validation purposes, rather than processing the entire model output comprehensively. By selectively extracting and validating only the location coordinates and corresponding document regions, the system reduces computational overhead and energy consumption while maintaining extraction efficiency and accuracy.
Data Source
AI summary
Systems and methods for performing document entity extraction are described herein. The method can include receiving an inference document and a target schema. The method can also include generating one or more document inputs from the inference document and one or more schema inputs from the target schema. The method can further include, for each combination of the document input and schema input, obtaining one or more extraction inputs by generating a respective extraction input based on the combination, providing the respective extraction input to the machine-learned model, and receiving a respective output of the machine-learned model based on the respective extraction. The method can also include validating the extracted entity data based on reference spatial locations and inference spatial locations and outputting the validated extracted entity data.