Value-Based Transaction Labeling Using ML Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual labeling of large-scale unlabeled transaction data is time-consuming and labor-intensive, making it inefficient for subcategory-level transaction classification and prioritization in procurement processes.
Innovation Solution
A computer-implemented method using a heuristics-based algorithm and machine learning models to rank and label unlabeled candidate transaction data, determining similarity scores and predicting subcategory labels to prioritize manual labeling efforts effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is applied to all unlabeled transaction data, then labeling accuracy is improved, but time consumption and labor intensity increase significantly
Solution Approach 1:
The system enables self-service by using automatically generated subcategory labels from the text classification model to label the majority of transactions, reducing reliance on manual labeling. Only high-value unlabeled transactions identified by the ranking algorithm require manual labeling, allowing the system to serve itself for the bulk of data processing while minimizing human intervention to essential cases only.
2Reliability
If manual labeling is applied to all unlabeled transaction data, then model training data quality is improved, but resource requirements increase significantly
Solution Approach 1:
The system applies local quality by differentiating between high-value and low-value transactions, applying manual labeling only to high-value transactions that significantly impact model performance. The text classification model automatically handles low-value transactions with acceptable accuracy, allocating human resources locally to where they provide maximum benefit rather than uniformly across all data.
Solution Approach 2:
The system changes the parameter of data quality by using the ranking algorithm to identify transactions with the largest positive impact on predictive model performance. This parameter-based selection ensures that manual labeling efforts are directed toward transactions that will most improve model training data quality, optimizing the relationship between resource investment and model improvement.
3Measurement precision
If subcategory-level classification is implemented, then classification precision is improved, but processing complexity increases
Solution Approach 1:
The system performs preliminary action by generating subcategory labels automatically using the text classification model before manual labeling is applied. This preliminary automatic labeling provides a foundation that reduces the burden of manual labeling and allows the system to achieve high classification precision by combining automated subcategory prediction with targeted manual verification of high-value transactions.
Data Source
AI summary
In one embodiment, a method includes assessing unlabeled transaction data and unlabeled candidate transaction data. The method includes determining a first plurality of features for the unlabeled transaction data and a second plurality of features for the unlabeled candidate transaction data using a Bag of Words model. The method includes determining matching pairs using the unlabeled transaction data and the unlabeled candidate transaction data. The method includes calculating similarity scores for the matching pairs using the first plurality of features and the second plurality of features. The method includes generating a machine learning model to predict a subcategory label and a ranking value for the unlabeled candidate transaction data using the similarity scores of the matching pairs. The method includes determining, using the machine learning model, the subcategory label and the ranking value associated with the unlabeled candidate transaction data.