Content-aware resolution sequence mining for ticket routing

Ticket routing is key to the efficiency of IT problem management. Due to the complexity of many reported problems, problem tickets typically need to be routed among various expert groups, to search for the right resolver. In this paper, we study the problem of using historical ticket data to make smarter routing recommendations for new tickets, so as to improve the efficiency of ticket routing, in terms of the Mean number of Steps To Resolve (MSTR) a ticket. Previous studies on this problem have been focusing on mining ticket resolution sequences to generate more informed routing recommendations. In this work, we enhance the existing sequence-only approach by further mining the text content of tickets. Through extensive studies on real-world problem tickets, we find that neither resolution sequence nor ticket content alone is sufficient to deliver the most reduction in MSTR, while a hybrid approach that mines resolution sequences in a content-aware manner proves to be the most effective. We therefore propose such an approach that first analyzes the content of a new ticket and identifies a set of semantically relevant tickets, and then creates a weighted Markov model from the resolution sequences of these tickets to generate routing recommendations. Our experiments show that the proposed approach achieves significantly better results than both sequence-only and content-only solutions.