Effect of Hidden Vector on the Speech of PRVA.

In this paper, we aim to suggest a novel method for designing a product recommendation virtual agent (PRVA) that can keep users motivated to interact with the agent. In prior papers, many methods of keeping users motivated postulated real-time and multi-modal interaction. We propose a novel method that can be used in one-direction interaction. We define the notion of the ``hidden vector," that is, information that is not mentioned by a PRVA and that the user can suppose spontaneously. We conducted an experiment to verify the hypothesis that PRVAs having a hidden vector are more effective than other PRVAs. As a result, this hypothesis was supported. Also, we conducted a second experiment to judge whether the effect observed in the first experiment was caused by the hidden vector or the length of speech. As a result, it was shown that the effect was caused by the hidden vector, not by the length of speech. From these results, our proposed method was shown to be effective.