Graph Neural Network Knowledge Graph for IT Incident Engineer Assignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional machine learning approaches for identifying ideal engineers to resolve IT service incidents are inefficient due to their inability to capture relationships among engineers who previously collaborated, leading to low accuracy and prolonged ticket resolution times.
Innovation Solution
A knowledge graph is developed using a graph neural network, incorporating historical incident data, user interactions, and product documentation to rank users best suited to address new incidents, leveraging natural language understanding and graph neural network theory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning approaches are used to identify engineers for resolving IT incidents, then the system is simple to implement, but the accuracy of engineer assignment is low and ticket resolution time is prolonged
Solution Approach 1:
The patent transitions from traditional machine learning approaches to graph neural networks, adding the dimension of relational structure to the engineer-incident matching system. By representing engineers, incidents, and their relationships as nodes and edges in a knowledge graph, the system captures collaborative relationships and expertise patterns that traditional flat ML models cannot detect, thereby improving assignment accuracy while managing complexity through structured representation.
Solution Approach 2:
The system changes the parameter representation from simple engineer profiles to enriched vectors that incorporate collaboration history, expertise domains, and incident context. The graph neural network dynamically updates engineer embeddings based on their relationships and past performance, transforming static parameters into adaptive, context-aware features that improve matching precision without linearly increasing system complexity.
2Reliability
If traditional machine learning models are used, then the model complexity is low, but the ability to capture relationships among engineers is insufficient
Solution Approach 1:
The patent introduces a knowledge graph as an intermediary structure between raw data and the machine learning model. This knowledge graph serves as a mediator that explicitly encodes relationships among engineers, incidents, and expertise domains, allowing the graph neural network to process relational information effectively. The intermediary structure enables reliable relationship capture while keeping the core ML model architecture manageable through structured input representation.
3Productivity
If the first engineer is assigned to solve a ticket, then the assignment process is quick, but the engineer may not be the final processor and resolution time increases
Solution Approach 1:
The system performs preliminary analysis using the graph neural network to predict the most suitable engineer for each incident before assignment. By pre-computing engineer embeddings that reflect their expertise, collaboration patterns, and historical performance, and by pre-ranking candidates based on incident characteristics, the system ensures that the first assigned engineer is highly likely to be the effective resolver, reducing iterative transfers and overall resolution time while maintaining efficient assignment processes.
Data Source
AI summary
In some embodiments, a computer system may compute a knowledge graph using a graph neural network and source data corresponding to historical incidents, with the source data comprising knowledge base article data, historical incident data, component data, user data, and swarm data. The computer system may compute a new incident vector based on new incident data using a natural language understanding algorithm, and, for each one of a plurality of users, compute an updated user vector using the knowledge graph. The computer system may then compute a ranked list of the plurality of users based on a comparison of the new incident vector with the corresponding updated user vector of each one of the plurality of users.