Document Information Extraction Using LLM Reference Prompts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical character recognition (OCR) technologies struggle to accurately extract character strings corresponding to a specified item when the item name in the image data differs from the expected item name, leading to inaccuracies in information extraction.
Innovation Solution
An information processing apparatus utilizing a large language model (LLM) with few-shot learning, incorporating document information and extraction results from similar reference documents to enhance accuracy by inputting prompts that include document information and extraction results from past documents, effectively leveraging tacit and business knowledge.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCR processing is used to extract character strings, then the extraction process is simple and fast, but the accuracy decreases when item names differ from expected names or are not included in the image data
Solution Approach 1:
The system performs preliminary actions by selecting and storing extraction results from reference documents before processing the target document. The selection unit identifies reference documents based on similarity to the target document, and the storage unit pre-stores their extraction results, which are then utilized to improve accuracy in the inference process.
Solution Approach 2:
The system introduces an intermediary mechanism by using reference documents and their extraction results as mediators between the target document and the final extraction outcome. The inference unit leverages these intermediary reference results to guide and improve the extraction accuracy for the target document, especially when item names differ or are missing.
2Measurement precision
If reference documents from past processing are utilized through few-shot learning, then extraction accuracy for items with different or missing names improves, but processing time and computational resources increase
Solution Approach 1:
The system applies partial action by selectively utilizing only relevant reference documents and their extraction results that are most similar to the target document, rather than processing all available reference data. The selection unit filters reference documents based on similarity criteria, and the inference unit uses only the necessary portion of stored extraction results, thereby reducing processing time while maintaining accuracy improvements.
3Productivity
If only the target document is processed without reference to past documents, then processing is faster and requires fewer resources, but accuracy decreases when item names are missing or differ from expected names
Solution Approach 1:
The system implements multi-functionality by enabling the information processing apparatus to perform both rapid single-document processing and enhanced accuracy processing with reference documents. The selection unit and storage unit allow the system to universally handle different processing scenarios, switching between using only the target document or incorporating reference documents based on the specific extraction challenge.
Data Source
AI summary
An information processing apparatus comprises: an acquisitor to obtain document information, including character strings and position data from document image data; a converter to transform the acquired document information into a distributed representation; an information extractor to identify a character string corresponding to an item specified by a prompt, using a large language model by inputting the prompt with the acquired document information; a storage unit to save the document information, distributed representation, and extraction results, associating them with the document; and a selector to choose a reference document from previously processed documents based on distributed representations. The prompt for processing a new document includes details about the selected reference document.