ML Classifiers for User Submission Field Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Incorrect field values in user submissions, such as mischaracterization of bug submissions or incorrect priority and assignee fields, lead to delays in processing and resolution in tracking systems, particularly in systems handling large volumes of submissions.
Innovation Solution
Implementing machine learning classifier modules to predict correct field values for tracking data structures, which compare predicted values with stored values to generate mismatch information and automatically update incorrect fields, thereby reducing the time to resolution (TTR) and improving user experience and system efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review of user submissions is performed to ensure field accuracy, then measurement precision of field values is improved, but productivity of the tracking system deteriorates due to increased processing time
Solution Approach 1:
The system performs preliminary validation of field values using machine learning classifiers before submissions are processed further. The classifiers predict whether field values are correct or incorrect, and flag potentially incorrect values for review, performing the accuracy check in advance rather than during manual processing
Solution Approach 2:
The tracking system automatically validates and flags incorrect field values using trained machine learning classifiers without requiring manual intervention for every submission. The system serves itself by identifying and correcting common field errors automatically, reserving manual review only for cases where the classifier uncertainty is high
2Reliability
If all user submissions are manually reviewed for field correctness, then reliability of processing is improved, but loss of time increases due to extensive manual intervention
Solution Approach 1:
The system automatically detects and flags incorrect field values using machine learning classifiers, reducing reliance on manual review. The classifiers independently assess field correctness and prioritize submissions for manual review based on predicted error probability, allowing the system to self-manage quality control
Solution Approach 2:
The system uses feedback from the machine learning classifiers to dynamically prioritize submissions for manual review. Submissions with high predicted error probability are routed for immediate review, while those with low probability proceed automatically, creating a feedback-driven processing workflow that optimizes both reliability and speed
3Productivity
If machine learning classifiers are implemented to automatically detect incorrect field values, then productivity is improved by reducing manual review needs, but device complexity increases due to additional processing components
Solution Approach 1:
The machine learning classifiers are trained in advance on historical submission data to learn patterns of incorrect field values. This preliminary training phase allows the system to perform accurate predictions during operation without requiring complex real-time analysis, as the classification logic has already been established beforehand
Solution Approach 2:
The machine learning classifiers act as an intermediary layer between submission intake and manual review processes. Rather than directly complex manual review workflows, the classifiers provide a simple automated filtering mechanism that identifies problematic submissions, simplifying the overall system architecture while maintaining high productivity
Data Source
AI summary
Techniques are disclosed relating to detecting one or more incorrect fields in user submissions, using machine learning techniques. A corrective system may access information for a plurality of fields of a tracking data structure for a user submission. In some embodiments, the corrective system predicts correct values for multiple fields of the tracking data structure using a plurality of respective different machine learning classifier modules. In some embodiments, the classifier modules use different sets of the plurality of fields as inputs and the multiple fields include a priority of the user submission and an assignee for the user submission. In some embodiments, in response to determining that at least one of the predicted correct values does not match a current value for a corresponding field of the tracking data structure, the computing system stores information indicating the mismatch and may automatically correct one or more fields.