Spoken Dialogue System Based on Attribute Vector for Travel Agent Robot

TL;DR

This paper presents a dialogue system for a travel robot that uses attribute vectors to understand and recommend sightseeing spots, with experiments showing the importance of correct understanding for user satisfaction.

Contribution

Introduces an attribute vector-based dialogue system for travel robots, enhancing recommendation accuracy and user interaction understanding.

Findings

Satisfaction increases with correct understanding and appropriate responses.

Negative correlation between number of user utterances and satisfaction.

System achieved a satisfaction score of 40.1 out of 63.

Abstract

In this study, we develop a dialogue system for a dialogue robot competition. In the system, the characteristics of sightseeing spots are expressed as "attribute vectors" in advance, and the user is questioned on the different attributes of the two candidate spots. Consequently, the system can make recommendations based on user intentions. A dialogue experiment is conducted during a preliminary round of competition. The overall satisfaction score obtained is 40.1 out of 63 points, which is a reasonable result. Analysis of the relationship between the system behavior and satisfaction scores reveals that satisfaction increases when the system correctly understands the user intention and responds appropriately. However, a negative correlation is observed between the number of user utterances and the satisfaction score. This implies that inappropriate responses reduce the usefulness of the system as a consultation partner.