DeepRouting: A Deep Neural Network Approach for Ticket Routing in Expert Network

Ticket routing is a part of software support process, where multiple expert groups are involved in processing incident tickets. The goal of routing is to find an expert group which can resolve a ticket at the initial assignment, or when it needs to be transferred to another group. Matching a ticket to its potential resolver effectively provides significant business value for both service providers and their customers. Previous works used hand-crafted features to train predictive models to automate or assist in routing. One of the findings shows that, the similarity between a ticket and an expert group is prominent in identifying the resolver among other groups. Meanwhile, numerous studies demonstrate the effectiveness of deep neural networks in text similarity modeling problems. In this paper, we propose a multi-view deep neural network solution to jointly learn a relevance score for a ticket-group pair, using both text and routing path information. The text relevance is modeled by a classic deep semantic matching model, while the routing graph representation is embedded using a convolutional graph network. Experimental results show that the proposed approach outperforms baseline models in resolver ranking and assistive routing tasks. Comparative experiments also show that text has higher importance than routing path information.