Application Object Labeling for Real-Time ML Data Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing software applications face challenges in integrating machine learning (ML) and artificial intelligence (AI) due to the time-consuming and resource-intensive process of adding labels during data preparation, leading to delays between data updates and model usage, especially in SaaS environments where upgrades require manual adjustments.
Innovation Solution
A Label Management System (LMS) that enables in-application, real-time labeling of data objects and their related objects, allowing users to extend and enrich data with custom labels, and automates label computation using rules and large language models, ensuring minimal re-labeling and lifecycle stability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data is labeled during data preparation pipeline, then ML model training can be performed, but time delay between data updates and model usage increases
Solution Approach 1:
The system performs preliminary labeling actions by adding label extension fields to application objects in real-time as data is created or modified. Labels are computed and stored proactively before ML model training is needed, eliminating the need for time-consuming batch labeling processes later. This allows ML models to immediately consume freshly labeled data without waiting for separate data preparation cycles.
Solution Approach 2:
The labeling process transitions from discrete batch operations to continuous real-time labeling. As application objects are created, updated, or deleted, labels are continuously computed and maintained through event triggers. This continuous labeling ensures that ML models always have access to current labeled data, eliminating gaps between data updates and model training opportunities.
2Adaptability or versatility
If manual labeling is performed during data engineering, then custom labels can be added, but resource consumption and time requirements increase
Solution Approach 1:
The system enables self-service labeling where application objects automatically generate their own labels through predefined rules and algorithms. When objects are created or updated, triggering events automatically invoke labeling logic without requiring manual data engineering intervention. This self-labeling capability maintains high adaptability for custom labels while dramatically improving productivity by eliminating manual labeling efforts.
Solution Approach 2:
The labeling functionality is merged directly into the application database and object model. Label extension fields are integrated as part of the application object structure, and labeling rules are combined with business logic. This merging eliminates the need for separate data engineering pipelines, allowing custom labels to be created and maintained within the application itself, thereby improving efficiency while preserving versatility.
3Ease of operation
If data is exported for labeling, then labeling can be performed, but the process becomes time-consuming and resource intensive
Solution Approach 1:
The system introduces label extension fields as intermediaries between application data and ML model requirements. These extension fields serve as a mediator layer that stores label information directly within the application database, eliminating the need to export data for labeling. The intermediary fields allow labeling operations to occur in-place within the application, improving accessibility while reducing time loss.
Solution Approach 2:
The label extension field mechanism provides universal labeling capability that works across all application objects and types. A single standardized approach using extension fields and mapping tables enables labeling for diverse object types without requiring object-specific export processes. This multi-functional labeling system improves ease of operation while minimizing data preparation time by handling all labeling uniformly within the application.
4Measurement precision
If labeling rules are changed, then label accuracy can be improved, but all instances must be reprocessed
Solution Approach 1:
The system segments the relabeling process by identifying and processing only affected instances when labeling rules change. Instead of reprocessing all data instances, the system determines which specific objects are impacted by rule changes and targets relabeling efforts only to those segments. This selective relabeling maintains label accuracy improvements while preserving productivity by avoiding unnecessary reprocessing of unchanged data.
Solution Approach 2:
The labeling system dynamically adapts to rule changes by implementing event-driven relabeling. When labeling rules are modified, the system dynamically identifies affected objects through event triggers and activates relabeling only for those instances. This dynamic approach allows label accuracy to be improved through rule changes while maintaining high relabeling efficiency by avoiding static full-reprocessing requirements.
Data Source
AI summary
A computer-implemented method includes adding, using a label management system (LMS), an object-specific multi-value label extension field to an application object (AO) associated with a software application. Using the LMS and to an AO user interface (UI) associated with the AO, a multi-value label UI extension field is added. Using the LMS, a label value table and associated label value maintenance UI is created to maintain label values. Using the LMS, a mapping table is created between the object-specific multi-value label extension field of the AO and the label value table. Using the LMS and the AO UI, label values assigned to the AO are read using foreign key retrieval from the mapping table and the label value table.