In this paper we give an overview of the evaluation metrics used to measure the performance of backchannel prediction models. Both objective and subjective evaluation metrics are discussed. The survey shows that almost every backchannel prediction model is evaluated with a different evaluation metric. This makes comparison between developed models unreliable, even beside the other variables in play, such as different corpora, language, conversational setting, amount of data and/or denition of the term backchannel.
|Publisher||University of Texas|
|Conference||Interdisciplinary Workshop on Feedback Behaviors in Dialog, Stevenson, Washington, USA|
|Period||7/09/12 → …|
- HMI-IA: Intelligent Agents