Document Information Extraction Using Referral Character Strings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing key-value extraction technologies fail to extract target information when it is not located near the target character string in a document, as seen in documents like contracts where attribute information is recorded separately from the key.
Innovation Solution
An information processing device that acquires a document image, identifies a target character string, and extracts corresponding information from a region set by the position of a related character string, using a referral character string or a relationship dictionary to locate the information even if it is not directly adjacent to the key.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If key-value extraction technology extracts target information only from regions near the target character string, then extraction simplicity is maintained, but extraction success rate deteriorates when target information is located far from the key
Solution Approach 1:
The patent introduces a referral character string as an intermediary element between the target character string (key) and the target information (value). When the value is not found near the key, the system searches for a referral character string that provides a reference to the location of the value, enabling indirect access to the target information while maintaining the key-value extraction framework
Solution Approach 2:
The patent extends the extraction process from a single-dimension approach (searching only near the key) to a multi-dimension approach by utilizing referral character strings that can point to different locations in the document. This adds a reference dimension to the extraction process, allowing the system to locate values that are spatially separated from their corresponding keys
2Productivity
If the extraction region is limited to areas near the target character string, then processing speed is maintained, but information completeness deteriorates when target information is recorded elsewhere
Solution Approach 1:
The patent performs preliminary extraction of the referral character string immediately after identifying the key, before attempting to locate the value. This preliminary action enables the system to quickly determine whether the value is nearby or referenced elsewhere, allowing for efficient branching of the extraction process without exhaustive searching
Solution Approach 2:
The referral character string serves as a mediator that bridges the gap between the key and the remotely located value. By extracting and processing this intermediary element, the system can efficiently navigate to the correct location of the target information without sacrificing processing speed
Data Source
AI summary
An information processing device includes a processor configured to acquire a document image illustrating a document, acquire a related character string associated with a target character string included in the document image, and extract target information corresponding to the target character string from a region set with reference to a position of the related character string in the document image.