Mining software support tickets for assistive routing

• Main Author: Han, Jianglei
• Other Authors: Sun Aixin
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Computer science and engineering::Software::Software engineering
• Online Access: https://hdl.handle.net/10356/143908
• Institution: Nanyang Technological University

The technical support organization of a software service provider consists of groups of experts who are processing customer tickets, which are electronic reports of any technical incident. Operationally, assigning a ticket to the correct expert is an important step towards resolution. An incorrect assignment would result in longer turnaround time and customer dissatisfaction, incurring cost and overhead. Ticket routing problem is about finding the right expert group that is the most capable of resolving a given ticket. In the ticket processing workflow, routing takes place in both initial assignment and inter-group transfer. Industrial and academic researchers developed and published automated solutions to analyze and process tickets, using different statistical and algorithmic methods. However, a good number of the previous works focus on either initial assignment or inter-group transfer, and propose solutions accordingly. Some take a multi-stage approach, combining the routing results sequentially, without leveraging synergies between the stages. Moreover, each stage only uses a subset of the information available. With more archived data and enhanced technologies, it calls for a fresh look and unified approach to the problem. In this thesis, we mine tickets from a software support system and investigate ticket routing problem in terms of routing performance evaluation, content analysis, and assistive routing. Firstly, we review and discuss the limitations of existing evaluation metrics and frameworks of routing systems, proposing a novel metric and the assistive routing evaluation framework. Next, we analyze tickets text and extract software product mentions, using manually crafted features. Combining insights from data mining and domain knowledge, we propose two routing frameworks based on learning-to-rank paradigm, incorporating features from ticket, group, ticketgroup, and group-group feature groups. Lastly, we present a deep neural network framework, using classic semantic networks to model ticket-group interaction. In assistive routing evaluation, our proposed frameworks are more effective compared to baseline routing systems using multi-stage approach.