Predicting help desk ticket reassignments with graph convolutional networks

Efficient triaging of incident tickets is a critical task in Information Technology Service Management. Introducing interventional measures on tickets that are difficult to resolve can help improve the triaging of complex tickets. This work reports a method to predict the resolution complexity of a reported incident. The number of times a ticket is reassigned is a measure of difficulty in resolving the incident. Ticket resolution is associated with a variable workflow. A graph representation of ticket resolution offers advantages from the standpoint of running ad hoc queries. Predicting ticket reassignments requires the application of machine learning to this graph. A Relational Graph Convolutional Network is used for this purpose. The developed model provides benefits beyond predicting ticket reassignments accurately. It provides embeddings that can be used to derive insights about the operation of the help desk organization and the users of the help desk. •Help desk tickets have a variable resolution path.•Graphs are a suitable data structure to capture this variable resolution path.•Number of analysts required for ticket resolution is a measure of ticket complexity•A Heterogeneous Graph Convolutional Network can be used to predict ticket complexity•Insights about ticket resolution can be obtained from graph embeddings.