User Modeling on Temporal Changes in User Behaviors for Spoken Dialogue Systems

我们通过关注通过自动语音识别(ASR)准确率和插入率衡量的用户行为的时间变化，对用户对口语对话系统的习惯进行了定量建模。这一改变用户行为的模型可以作为实现用户自适应交互的基础发现。该模型适用于产生自适应响应并从可能包含错误的ASR结果中选择正确ASR结果的口语对话系统。通过对插入话语选择准确率的提高，验证了该方法的有效性。

We quantitatively modeled users' habituation to a spoken dialogue system by focusing on temporal changes in user behaviors measured by automatic speech recognition (ASR) accuracy and barge-in rates. This model on changing user behaviors can be used as a fundamental finding for realizing user-adaptive interactions. The model is applicable to spoken dialogue systems that generate adaptive responses and select correct ASR results from those that may contain errors. We showed its effectiveness by the improvement of selection accuracy for barge-in utterances.