Automated Training Data Generation for AI Incident Resolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual assembly of large training datasets required for neural networks is a time-intensive and complex task, especially in specialized fields like IT incident management, where data quality and completeness vary significantly due to human entry, making it impractical for human review and limiting the adoption of AI technologies.
Innovation Solution
An automated system that collects and correlates incident and resolution data from multiple sources, including IT environments, runbooks, and user interactions, to create a knowledge database used for training machine learning models, which then provides resolution recommendations and updates based on user feedback.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual assembly of training datasets is used, then data quality and completeness can be controlled, but the process is time-intensive and complex
Solution Approach 1:
The system automatically collects, correlates, and stores incident and resolution data from multiple sources without requiring manual assembly. The automated data collection process queries databases, monitors systems, and aggregates relevant information to create training datasets independently, eliminating the time-consuming manual curation process while maintaining data quality through structured collection protocols.
Solution Approach 2:
The patent replaces manual data assembly operations with automated computational processes. Instead of humans manually collecting and verifying data, the system uses automated queries, data correlation algorithms, and database operations to assemble training datasets, substituting mechanical manual labor with digital automation.
2Productivity
If automated data collection is used, then productivity increases, but data quality and completeness may deteriorate
Solution Approach 1:
The system incorporates feedback mechanisms where collected data is validated against expected patterns and quality criteria. The automated collection process continuously monitors data quality metrics and can adjust collection parameters or trigger re-collection if quality thresholds are not met, ensuring high data quality while maintaining automated high-speed collection.
Solution Approach 2:
The system performs preliminary data validation and correlation operations during the automated collection process itself, rather than requiring separate manual review stages. Data is pre-filtered, correlated with relevant information, and validated against quality criteria before being stored as training data, ensuring quality is built-in during automated collection.
3Reliability
If human review of data is performed, then data completeness can be verified, but the complexity and time required increase significantly
Solution Approach 1:
The data processing system is divided into distinct modular components: data collection modules for querying different sources, data correlation modules for linking incident and resolution data, quality validation modules for verifying completeness, and storage modules for organizing training datasets. This segmentation allows each component to be optimized independently and simplifies the overall complex processing pipeline.
Solution Approach 2:
The automated system performs multiple functions that would traditionally require separate manual processes: collecting data from multiple sources, correlating incident and resolution information, validating data quality, and preparing training datasets. A single automated pipeline executes all these functions sequentially, reducing overall complexity compared to manual multi-step verification processes.
Data Source
AI summary
An embodiment includes detecting incident data and resolution data in monitored data collected while monitoring an information technology (IT) environment. The embodiment correlates the incident data with the resolution data according to a detected change in health metrics data from the monitored data. The embodiment stores the correlated incident data and resolution data as a training dataset stored in a database and then trains a machine learning model using the training dataset. The embodiment deploys the trained machine learning model such that the trained machine learning model provides resolution recommendation in response to receiving new incident data.