Which Text-Mining Technique Would Detect Most Accurate User Frustration in Chats With Conversational Agents?
Hauke Hinrichs, Nguyen-Thinh Le · Electronic workshops in computing · 2018
Conversational agents are getting ever more prevalent in online activities. There are many different approaches to measuring acceptance rate for such systems. In this paper, we explore the option of detecting user frustration in text-based user messages. Five text mining techniques (Decision Table Majority, Naive Bayes, Multilayer Perceptron, Sequential minimal optimisation, and K*) are compared in a supervised learning scenario using different quantifiable parameters. The comparison between these techniques shows that Sequential Minimal Optimisation is quickest and most accurate for detecting user frustration in text-based user messages.