How angry are your customers? Sentiment analysis of support tickets that escalate

TL;DR

This paper investigates the use of sentiment analysis on support ticket conversations to identify indicators of potential escalations, aiming to improve prediction and handling of costly support issues.

Contribution

It demonstrates that sentiment analysis can distinguish escalated tickets from non-escalated ones, providing a foundation for predictive tools in support systems.

Findings

Significant sentiment differences between escalated and non-escalated tickets

Sentiment analysis shows potential as an indicator for escalation prediction

Provides groundwork for developing escalation prediction models

Abstract

Software support ticket escalations can be an extremely costly burden for software organizations all over the world. Consequently, there exists an interest in researching how to better enable support analysts to handle such escalations. In order to do so, we need to develop tools to reliably predict if, and when, a support ticket becomes a candidate for escalation. This paper explores the use of sentiment analysis tools on customer-support analyst conversations to find indicators of when a particular support ticket may be escalated. The results of this research indicate a considerable difference in the sentiment between escalated support tickets and non-escalated support tickets. Thus, this preliminary research provides us with the necessary information to further investigate how we can reliably predict support ticket escalations, and subsequently to provide insight to support analysts to better enable them to handle support tickets that may be escalated.