Composite ML System for Transaction Label Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models require high-quality, representative training data for accurate predictions, but gathering and verifying correctly labeled instances is time-consuming and difficult to automate, especially in domains with constant changes such as merchant names and tax classifications.
Innovation Solution
A composite machine learning system that uses explicit and implicit user feedback to determine labels for transactions, incorporating N-grams, associative arrays, and machine learning models to generate features and predict labels, with a feedback loop to update training data and improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models use traditional training data collection methods, then prediction accuracy can be improved, but the time and effort required for data gathering and verification increases significantly
Solution Approach 1:
The system enables automatic label verification through user feedback mechanisms. When users interact with predicted labels (e.g., correcting or confirming them), this feedback is automatically captured and used to verify and improve future predictions, eliminating the need for manual verification of each training instance
Solution Approach 2:
The system implements a feedback loop where user interactions with predicted labels (corrections, confirmations, or implicit feedback through usage patterns) are collected and used to continuously improve the machine learning model's prediction accuracy, creating a self-improving system that reduces manual intervention over time
2Measurement precision
If machine learning models are trained with more training instances, then predictive accuracy improves, but the complexity of managing and verifying training data increases
Solution Approach 1:
The system automatically manages training data quality through user feedback. Users implicitly verify or correct labels through their interactions, and the system automatically incorporates this feedback into the training dataset, eliminating the need for complex manual management and verification processes
Solution Approach 2:
The system introduces an intermediary feedback mechanism between the machine learning model and the training data. User feedback acts as a mediator that automatically verifies and refines training labels, simplifying the management of large training datasets without requiring direct human intervention in each instance
3Adaptability or versatility
If the system uses a composite machine learning approach with multiple models and feedback loops, then adaptability to domain changes improves, but system complexity increases
Solution Approach 1:
The composite system uses feedback loops from user interactions to continuously adapt to domain changes. When users correct or confirm labels in changing domains (e.g., new merchant names or tax classifications), this feedback automatically updates the training data, enabling the system to adapt to domain evolution without manual reconfiguration
Solution Approach 2:
The system employs a multi-functional composite approach where multiple machine learning models work together, each handling different aspects of label prediction. This universal architecture can adapt to various domains and changes by leveraging collective model performance and user feedback, rather than requiring domain-specific customizations
Data Source
AI summary
The present disclosure provides a composite machine learning system for a transaction labeling service. A transaction labeling service receives at least one descriptive string describing a transaction associated with a user. The service identifies a preliminary grouping from a generalized scheme. The service extracts a set of N-grams from the descriptive string and converts the N-grams and the preliminary grouping into a set of features. A machine learning model determines a label from a labeling scheme for the transaction based on the features. User input related to the label includes an accuracy indicator and a reliability indicator. If the reliability indicator satisfies a reliability condition, a set of training data for the machine learning model is updated based on the descriptive string and the label. The machine learning model is then trained against the updated set of training data.