Document Tagging via Numeric Token Replacement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for automatically tagging financial documents with XBRL tags, particularly for numeric and date values, deliver poor and unsatisfactory results due to the dynamic and unstructured nature of text notes.
Innovation Solution
A computer-implemented method that preprocesses electronic documents by extracting text, replacing numbers and dates with predetermined symbols, and tokenizing the text. A deep learning module then determines tags for the tokens, improving the accuracy of tagging numeric and date values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing automatic tagging methods are used for financial documents, then the tagging process is automated, but the accuracy of tagging numeric and date values is poor
Solution Approach 1:
The patent applies preliminary action by preprocessing the text before tagging - specifically replacing numbers and dates with special tokens before the main tagging process. This preparatory step ensures that numeric and date values are properly represented and can be accurately tagged by the deep learning model, resolving the accuracy issue while maintaining automation.
Solution Approach 2:
The patent changes the parameter representation of numeric and date values by replacing them with special tokens (e.g., <NUM1>, <DATE1>) that encode both the type and position information. This parameter transformation allows the deep learning model to better understand and accurately tag these values, improving measurement precision while keeping the system automated.
2Measurement precision
If manual tagging is performed by human auditors, then tagging accuracy can be maintained, but considerable time is spent reviewing each tag
Solution Approach 1:
The patent implements self-service by creating a system that automatically handles the tagging process without human intervention. The deep learning model with special token preprocessing autonomously tags numeric and date values accurately, eliminating the need for human auditors to manually review each tag while maintaining high accuracy levels.
Solution Approach 2:
The patent replaces the mechanical human auditing process with an automated deep learning system. The neural network model, enhanced with special token representation for numbers and dates, substitutes human reviewers and performs tagging automatically, significantly reducing time loss while maintaining or improving accuracy compared to manual methods.
3Device complexity
If standard tokenization is applied to text with numbers and dates, then text processing is simplified, but sub-word fragmentation occurs reducing tagging quality
Solution Approach 1:
The patent applies local quality by treating numbers and dates differently from regular text - replacing them with special tokens that preserve their integrity. This localized special handling prevents sub-word fragmentation of numeric and date values while keeping the rest of the text processing simple, thereby maintaining both low complexity and high tagging quality.
Solution Approach 2:
The patent introduces special tokens (e.g., <NUM1>, <DATE1>) as intermediaries between the original numeric/date values and the deep learning model. These intermediary tokens prevent fragmentation during tokenization while still conveying the essential information needed for accurate tagging, resolving the conflict between processing simplicity and tagging quality.
Data Source
AI summary
System and methods (100) for automatically tagging electronic documents are disclosed. An input module receives (102) an electronic document to be tagged. A preprocessing module then preprocesses (104) the electronic document to be tagged. The preprocessing of the electronic document comprises extracting a text from the electronic document to be tagged, replacing a number or a date in the extracted text with a predetermined symbol, and tokenizing the extracted text with the predetermined symbol into a plurality of tokens. After the preprocessing (104), a deep learning module determines (106) a tag for at least one of the plurality of tokens. The determined tag for the at least one token is then output (108) by an output module.