Invoice Account Code Recommendation Using NLP and Historical Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face inefficiencies in automatically recommending account codes for non-PO-backed invoices, requiring manual effort and extensive retraining of supervised models for each use case, which is resource-intensive and impractical.
Innovation Solution
A computer-implemented method using natural language processing and machine learning techniques to automatically detect and recommend account codes by filtering invoice data, identifying categories, and computing similarity scores based on historical invoices, reducing manual input and infrastructure burden.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a supervised model is trained and customized separately for each use case of computer-generated invoices, then the model can accurately predict account codes for that specific use case, but it requires significantly more CPU time, memory, storage, and network bandwidth, and involves extensive processing time and infrastructure resources
Solution Approach 1:
The patent applies a single universal supervised model that can handle multiple types of invoices (PO-backed and non-PO-backed) without requiring separate models for each use case. This universal model processes different invoice types through the same architecture, eliminating the need for multiple specialized models and significantly reducing computational resources, CPU time, and infrastructure requirements while maintaining accurate account code predictions across diverse invoice scenarios
2Use of energy by moving object
If a single supervised model is used for all invoice types, then infrastructure resources are reduced, but the model cannot be easily updated to incorporate new account codes without complete retraining
Solution Approach 1:
The patent implements a dynamic model update mechanism that allows the supervised model to be updated incrementally with new account codes and invoice data without requiring complete retraining. The model can adapt to new account codes by processing new examples through continuous learning mechanisms, enabling flexible adaptation to changing accounting requirements while maintaining the efficiency of a single unified model architecture
3Measurement precision
If manual entry of data is performed for non-PO-backed invoices, then accuracy can be controlled, but it requires tedious and labor-intensive effort from AP users
Solution Approach 1:
The patent implements an automated account code prediction system that performs self-service by automatically determining appropriate account codes for non-PO-backed invoices without requiring manual intervention from AP users. The system processes invoice data, predicts account codes using the supervised model, and presents recommendations automatically, eliminating tedious manual data entry while maintaining high accuracy through intelligent prediction algorithms
4Measurement precision
If extensive retraining of supervised models is performed for each use case, then model accuracy for specific invoice types improves, but processing time and infrastructure burden increase significantly
Solution Approach 1:
The patent employs a single universal supervised model that handles multiple invoice types including PO-backed and non-PO-backed invoices without requiring separate models for each use case. This universal approach eliminates the need for extensive retraining for each specific invoice type, significantly reducing model retraining time and infrastructure burden while maintaining high prediction accuracy across different invoice scenarios through a unified learning architecture
Data Source
AI summary
A computer-implemented method of detecting account codes and displaying the detected account codes on a graphical user interface comprising receiving, by a recommendation engine of a recommendation system, invoice data comprising supplier-customer information that corresponds to a supplier-customer transaction, wherein the invoice data comprises invoice descriptions and invoice characters, wherein the invoice descriptions and the invoice characters define contexts and patterns; determining, by the recommendation engine, that an amount of the invoice characters is not more than a preset threshold number of characters; in response to determining that the amount of the invoice characters is not more than the preset threshold number of characters, filtering, by the recommendation engine, the invoice descriptions of the invoice data based on predetermined constraints to extract filtered invoice data comprising filtered description lines and to generate a training corpus for a pre-trained Natural Language Processing (NLP) model; identifying, by classifying the filtered description lines with the pre-trained NLP model, one or more categories associated with each of the filtered description lines of the filtered invoice data; matching, by the recommendation engine, each of the identified one or more categories with one or more predefined historical categories, wherein the contexts and patterns associated with the filtered description lines are matched with predefined contexts and patterns of predefined historical invoice data that corresponds to the same supplier-customer information; generating, by the recommendation engine, a feature vector for the invoice data based on the matching; computing, by the recommendation engine, a categorical similarity score for each of the identified one or more categories based on the feature vector and an additional feature vector, wherein the additional feature vector is based on the predefined historical invoice data; and displaying, by the recommendation engine on the graphical user interface, a recommendation including an account code based on the computed categorical similarity score of each of the one or more categories to map the account code to the invoice data.