Financial Document Token Labeling for High-Confidence Datasets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic data extraction techniques for accounting documents, such as invoices or receipts, are unreliable in efficiently and accurately extracting relevant entity information like entity names and financial details.
Innovation Solution
A method involving token extraction, preliminary labeling using attribute labelling modules, and refined labeling with an expectation maximization model to determine confidence values for attribute labels, ensuring high accuracy in identifying entity attributes like ABN, BSB, and account numbers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automatic data extraction techniques are used for accounting documents, then data extraction speed is improved, but extraction accuracy deteriorates
Solution Approach 1:
The patent segments the data extraction process into multiple specialized attribute labelling modules, each responsible for specific attribute types (e.g., ABN detector, BSB detector, account number detector). This segmentation allows each module to focus on particular patterns, improving overall extraction accuracy while maintaining automated processing speed.
Solution Approach 2:
The patent introduces an expectation maximisation model as an intermediary that processes preliminary labels from multiple attribute labelling modules. This intermediary synthesizes conflicting or overlapping labels by calculating confidence values, thereby resolving accuracy issues that arise from automated extraction without requiring manual review of each document.
2Measurement precision
If multiple attribute labelling modules are used to improve label accuracy, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent merges the outputs of multiple attribute labelling modules into a unified expectation maximisation model. Instead of maintaining separate processing pipelines for each attribute type, the model consolidates preliminary labels from various detectors (ABN, BSB, account number, etc.) and resolves conflicts through a single confidence calculation process, reducing operational complexity.
Solution Approach 2:
The expectation maximisation model performs self-service by automatically resolving label conflicts without external intervention. It calculates confidence values for competing labels and selects the most appropriate attribute values autonomously, eliminating the need for manual disambiguation while managing the complexity introduced by multiple labelling modules.
3Reliability
If confidence values are calculated for all attribute labels, then reliability is improved, but computational time increases
Solution Approach 1:
The patent applies partial action by calculating confidence values selectively rather than for all possible attribute labels uniformly. The expectation maximisation model focuses computational effort on resolving conflicts where multiple labels compete for the same attribute, skipping confidence calculations for unambiguous cases, thereby reducing overall computational time while maintaining reliability.
Data Source
AI summary
A method comprises determining token(s) from financial documents and determining a set of preliminary attribute labels for the token(s), wherein the set is associated with attribute type(s). The method further comprises providing the set for each token to an attribute prediction model to determine, for the token, a confidence value for each attribute type(s), determining subsets of token, each subset being associated with a respective document of the plurality of documents and determining a set of refined labels for each document based on the confidence values, wherein the set of refined labels comprises a value for attribute type(s).