Invoice Processing ML Model Using Unlikely Combination Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated text processing methods, particularly for invoices, are computationally inefficient due to the large number of variables and complex relationships, making it difficult to accurately and quickly analyze and process invoices using machine-learning models.
Innovation Solution
A system utilizing artificial intelligence techniques, including natural language processing and machine learning, to extract structured information from unstructured narratives, generate multi-class classifications, and dynamically adapt models using genetic algorithms to improve prediction accuracy and speed, while identifying irregularities and non-compliant submissions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine-learning models are used to process invoices with many variables and complex relationships, then prediction accuracy can be improved, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent segments the invoice processing task into multiple stages: extracting features from unstructured text, generating structured representations, training separate ML models for different aspects (e.g., line item validation, invoice-level predictions), and combining results. This segmentation reduces the complexity of any single model while maintaining overall prediction accuracy.
Solution Approach 2:
The system performs preliminary processing of invoice data by extracting structured features from unstructured text before feeding them to ML models. This preliminary action prepares the data in a format that reduces computational complexity during the actual prediction phase, as models work with pre-processed features rather than raw unstructured text.
2Measurement precision
If more features and dimensions are included in the ML model to capture complex invoice relationships, then prediction accuracy improves, but processing speed and efficiency decrease
Solution Approach 1:
The patent extracts only the most relevant features from invoice data using natural language processing and text analysis techniques. By taking out and selecting only the critical features needed for prediction, the system maintains high prediction accuracy while reducing the dimensionality of input data, thereby improving processing speed.
Solution Approach 2:
The system applies different processing strategies to different parts of the invoice data based on their importance and characteristics. Critical fields receive more sophisticated analysis while less important fields are processed more simply, optimizing the balance between accuracy and speed for each specific feature.
3Productivity
If rule-based processing is used to handle structured invoice data, then processing speed is maintained, but accuracy decreases due to inability to handle complex relationships and deviations from common practices
Solution Approach 1:
The patent merges rule-based processing and machine-learning approaches into a hybrid system. Rules handle straightforward, structured invoice elements quickly, while ML models handle complex relationships and exceptions. This combination maintains processing speed for routine items while improving accuracy for complex cases through collaborative processing.
Data Source
AI summary
A description of a machine learning (ML) model is received, with the ML model including multiple features such as an unlikely combination feature, which corresponds to a first attribute to be located in an invoice and a second attribute to be located the invoice concurrently with the first attribute. Training data is received, including (i) invoice data with multiple invoices, each including the first attribute and the second attribute, and respective values of the first attribute and the second attribute, and (ii) validity data including indications of which of the invoices are valid and which of the invoices are invalid. The ML model is trained using the training data using the ML model. The training includes applying the values of the attributes to the unlikely combination feature. The ML model is applied to an invoice to be validated to determine a probability that the invoice is invalid.