Domain-Specific Generative ML Models for IT Incident Resolution

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

General-purpose machine learning models face challenges in providing accurate and efficient solutions for specific domains like IT incident handling due to inadequate training data quality and resource-intensive training processes, often resulting in inaccurate or incomplete resolutions.

Innovation Solution

A system that processes domain-specific incident tickets to generate high-quality training data by using a domain-specific statement classifier, ticket processor, and resolution quality evaluator to refine and retain accurate resolutions, then trains a domain-specific machine learning model for efficient incident resolution generation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If general-purpose ML models are applied to domain-specific tasks, then wide applicability is achieved, but accuracy in specific domains deteriorates

Engineering Contradiction:
Improvewide applicabilityVSAvoidaccuracy in specific domain
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent segments the training data into domain-specific subsets and trains separate ML models for different domains (e.g., IT incident handling, healthcare). This allows each model to specialize in its specific domain, improving accuracy while maintaining overall system versatility through multiple domain-specific models.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces domain-specific preprocessing layers and data curators as intermediaries between raw data and the ML models. These intermediaries adapt general-purpose data into domain-specific formats, enabling models to achieve high domain accuracy while retaining broad applicability through the intermediary adaptation layer.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If general-purpose ML models are trained on available data, then training speed is improved, but data quality insufficiency worsens

Engineering Contradiction:
Improvetraining speedVSAvoiddata quality
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent performs preliminary data quality assessment, cleaning, and domain-specific preprocessing before training begins. This preliminary action ensures that only high-quality, domain-relevant data is used for training, improving reliability while maintaining training efficiency through pre-processed ready-to-use data.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements automated data quality evaluation metrics and self-cleaning mechanisms that allow the system to automatically identify and remove low-quality data points during the training process, maintaining data quality standards without requiring extensive manual intervention or slowing down training.

Inventive Principle:
Principle #25Self-service

3Reliability

If domain-specific processing is applied to training data, then data quality is improved, but processing complexity increases

Engineering Contradiction:
Improvetraining data qualityVSAvoidprocessing complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies domain-specific processing only to the relevant portions of training data for each particular domain, rather than uniformly processing all data. This local quality approach improves data quality for specific domains while minimizing unnecessary processing complexity in areas where simple processing suffices.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent dynamically adjusts processing parameters and thresholds based on the specific domain characteristics and data quality metrics. This allows the system to optimize the level of processing complexity applied to different data subsets, achieving high quality outcomes without uniformly high complexity across all processing operations.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4439332A1Domain-specific generative machine learning models
Publication Date: 2024.10.02 BMC HELIX INC
  • EP4439332A1 patent drawingFigure 1
  • EP4439332A1 patent drawingFigure 2
  • EP4439332A1 patent drawingFigure 3

AI summary

A plurality of resolved incident tickets may each include a worklog providing a history of actions taken during attempts to resolve a corresponding resolved incident and a resolution having at least one resolution statement. An iterative processing of the plurality of resolved incident tickets may include processing each resolution statement of the resolution with at least one domain-specific statement classifier specific to the incident domain to either discard or retain a classified resolution statement; processing each retained classified resolution statement in conjunction with the worklog to determine whether to discard or retain the resolved incident; providing an updated resolution for the resolved incident when the resolved incident is retained, and adding the resolved incident with the updated resolution to the processed incident tickets. Then, at least one machine learning model may be trained to process a new incident ticket, using the processed incident tickets.