EasyTicket: a ticket routing recommendation engine for enterprise problem resolution

Qihong Shao, Yi Chen, Shu Tao, Xifeng Yan, Nikos AnerousisPVLDB 2008 · DOI

TL;DR

This paper introduces EasyTicket, a system that uses historical ticket resolution data to build a "social network" of expert groups and recommend the most efficient routing for new problem tickets. By mining these sequences, EasyTicket significantly reduces the time it takes to resolve issues, improving overall IT service efficiency.

What Problem Does This Paper Address?

In large organizations, IT support departments handle thousands of problem tickets daily, ranging from simple login issues to complex application failures. The efficiency of resolving these problems hinges critically on ticket routing – the process of transferring a ticket among different expert groups until it reaches the right resolver. While many ticket management systems exist, the routing process is often manual, relying on the limited knowledge and experience of support personnel. This can lead to tickets being sent to incorrect groups, causing delays and customer dissatisfaction. The challenge is to automate and optimize this routing process to improve resolution efficiency.

Methods

The core of EasyTicket is a Markov model that captures the "social network" of expert groups based on historical ticket routing sequences.

  • Sequence Extraction: The system first extracts sequences of group transfers from historical resolved tickets. For example, a ticket might be routed through groups A, then B, then C.
  • Markov Model Construction: A hybrid Markov model is built for each problem category. Each group is treated as a state, and the transition probabilities between states represent the likelihood of a ticket being transferred from one group to another. The model's order (how many past states are considered) is determined by minimizing conditional entropy.
  • Variable-order Multiple active state Search (VMS) Algorithm: For a new, open ticket, the VMS algorithm uses the Markov model to recommend the next best group to route it to. It maintains a set of visited groups and candidate groups, iteratively selecting the group that maximizes the transfer probability, considering variable orders of the Markov model.

Datasets

The system was evaluated on a dataset of 1.4 million real-world problem tickets collected from IBM's problem management system between January 1, 2006, and December 31, 2006. This dataset spanned 553 problem categories. The data was split into training and testing sets to build the Markov models and evaluate the recommendation system.

Key Insights and Ideas

  • Enterprise Social Network for IT Support: The paper frames the relationships between expert groups in ticket resolution as a "social network." This perspective allows for modeling and leveraging these relationships for better routing.
  • Power of Routing Sequences: The authors demonstrate that simply mining the sequence of group transfers, without deeply analyzing the ticket content itself, can significantly improve routing efficiency.
  • Probabilistic Modeling for Uncertainty: The use of Markov models acknowledges that routing decisions are probabilistic. The system learns these probabilities from historical data to make informed recommendations.
  • Hybrid Approach for Categorization: Building separate Markov models for different problem categories is crucial because the relationships and expertise required vary significantly across problem types.

Key Results

EasyTicket demonstrated a significant improvement in ticket routing efficiency:

  • Reduced Mean Steps To Resolve (MSTR): The overall MSTR for all problem categories was reduced from 3.94 (human decisions) to 2.58, a 34.52% reduction.
  • Efficiency Gains in Specific Categories: Notable reductions were seen in categories like AIX (43.15%) and DB2 (46.23%).
  • Fast Model Building and Recommendation: The Markov model for each problem category could be built in less than 1 second, and generating a ticket routing recommendation took less than 1 ms per ticket.
  • Reduced Routing Sequences: In a case study, EasyTicket reduced the number of groups involved in resolving a sample ticket from nine to three, highlighting its ability to avoid inefficient detours.

Limitations and Future Work

  • Ticket Content Ignored: The current study focuses solely on routing sequences. The authors suggest that incorporating ticket content (problem descriptions, diagnostic data) using text mining techniques could further enhance the social network model and improve routing accuracy.
  • Broader Applications: The developed social network model has potential applications beyond ticket routing, such as discovering organizational issues, performance benchmarking, and improving enterprise problem management in general.
  • Dynamic Environments: While the paper mentions testing sensitivity to time-variability, future work could explore more adaptive models that continuously learn and update as new tickets arrive and group expertise evolves.

Summary updated

BibTeX
@article{shao2008easyticket,
  author = {Qihong Shao and Yi Chen and Shu Tao and Xifeng Yan and Nikos Anerousis},
  title = {{EasyTicket: a ticket routing recommendation engine for enterprise problem resolution}},
  journal = {Proc. VLDB Endow.},
  year = {2008},
  doi = {10.14778/1454159.1454193},
}
Figure