A comparison of three think-aloud protocols used to evaluate a voice intelligent agent that expresses emotions

ABSTRACT

This study proposes two adapted think-aloud protocols for the evaluation of a voice intelligent agent. In the adapted retrospective think-aloud (RTA) protocol, users verbalise their thoughts based on the chat history after task-completion. In the adapted interactive think-aloud (ITA) protocol, users verbalise their thoughts regarding the intelligent agent being evaluated without the help of a facilitator. This study compares these two protocols with the classical think-aloud protocol (CTA) for evaluating an intelligent agent in terms of task time and verbal utterances. The influence of the intelligent agent’s emotional expression is also considered. The results suggest RTA is suitable for collecting user experience and causal explanation of utterances, CTA for collecting recommendation and prediction utterances, and ITA for collecting problem formulation and recommendation utterances. Furthermore, CTA and RTA can collect more total utterances, while CTA and ITA are influenced less the VIA’s emotional expression. This study provides guidelines by which future evaluators can choose suitable think-aloud protocols.

KEYWORDS:

• Voice intelligent agent
• think-aloud protocol
• usability evaluation
• emoticon
• utterance category

Acknowledgement

This study was funded by a National Key Research and Development Plan grant 2016YFB1001200-2.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This study was funded by a National Key Research and Development Plan [grant number 2016YFB1001200-2].