A Dialogue Robot System to Improve Credibility in Sightseeing Spot Recommendations Naoki Yoshimaru1 Tomohiro Masuda2 Hyejin Hong2 Yusei Tanaka2 Motoharu Okuma2

2025-04-30 0 0 701.54KB 6 页 10玖币

侵权投诉

A Dialogue Robot System to Improve Credibility

in Sightseeing Spot Recommendations

Naoki Yoshimaru1, Tomohiro Masuda2, Hyejin Hong2, Yusei Tanaka2, Motoharu Okuma2,

Nagihiro Matsumoto2, Kazuma Kusu1, Takamasa Iio2, and Kenji Hatano2

Abstract— Various studies have been conducted on human-

supporting robot systems. These systems have been put to

practical use over the years and are now seen in our daily

lives. In particular, robots communicating smoothly with people

are expected to play an active role in customer service and

guidance. In this case, it is essential to determine whether the

customer is satisﬁed with the dialog robot or not. However, it

is not easy to satisfy all of the customer’s requests due to the

diversity of the customer’s speech. In this study, we developed a

dialog mechanism that prevents dialog breakdowns and keeps

the customer satisﬁed by providing multiple scenarios for the

robot to take control of the dialog. We tested it in a travel

destination recommendation task at a travel agency.

I. INTRODUCTION

A dialogue robot competition is being held to test mul-

timodal dialogue techniques using humanoid robots[1]. In

that dialog robot competition, a humanoid robot acts as a

counter-sales person for a travel agency and recommends

travel destinations to users. Participants compete in terms

of user satisfaction with the robot’s recommendations and

the naturalness of the dialogue. In order to improve user

satisfaction, it is necessary to not only present information

on travel destinations but also to listen to user requests and

recommend destinations that meet those requests.

It is not easy to listen to user requests through dialogue,

respond to those requests, and reﬂect those requests in the

recommendations. In particular, when users are allowed to

speak freely, classical rule-based dialogue systems often

break down. This is because the content of user utterances

is diverse, and it is impossible to prepare rules for all

utterances. The breakdown of dialogue can be avoided by

processing the occurrence of utterances that do not exist

in the rules(exception handling). However, frequent occur-

rences of exceptions may reduce user satisfaction. These

problems also arise in example-based dialogue systems.

Another approach is to use a generation-based dialog system.

Generative dialogue systems can generate natural responses

using large-scale language models and prompts. However,

for the speciﬁc task of recommending destinations, they may

generate information that is not true (fake information). For

example, the system may output information about exhibits

or products that do not exist in the destination description.

Since such fake information may cause ethical and legal

problems, adopting this method with current technology is

1Graduate School of Culture and Information Science, Doshisha Uni-

versity, 1-3 Tatara-Miyakodani Kyotanabe, Kyoto 610-0394, Japan

2Faculty of Culture and Information Science,Doshisha University, 1-3

Tatara-Miyakodani Kyotanabe, Kyoto 610-0394, Japan

not easy. Therefore, this study adopts a rule-based dialogue

system where the designer has complete control over the

utterances, although it is not as versatile as a generation-

based dialogue system.

In a rule-based dialog system, the robot must take the

initiative in the dialog to reﬂect the user’s requests in the

recommendations. Speciﬁcally, the robot takes the initiative

in the dialog robot by repeatedly asking the user questions.

Here, the robot’s questions should be questions that can be

answered with Yes/No or choices (choice-type questions).

For example, ”Do you travel alone? Or do you travel with

your family or friends?.” These questions should be designed

so that the user can answer them. Such questions implicitly

limit how users respond, making it easier to predict their

answers. As a result, it can reduce the possibility of dialogue

breakdown while still allowing the user to speak regarding

the destination recommendation.

However, in recommending travel destinations, the user is

not likely to be fully satisﬁed if the robot takes the initiative

in the dialog and asks many questions. Methods in which the

robot takes the initiative in dialog have been applied to daily

conversations with the elderly [2], and dialog with visitors

at events [3]. According to those studies, the method can

elicit some degree of user satisfaction. However, the dialogue

in those tasks is closer to chit-chat and different from

the dialogue in recommending travel destinations. Being

bombarded with questions by the robot in a chat is, in a

sense, easier for the user, who has no purpose in interacting

with the robot. This is because the user does not have to

think about what to say to the robot. On the other hand, in

recommending a travel destination, the user has the explicit

goal of selecting a destination. Therefore, it is important that

the robot’s questions are meant for destination selection and

that the user’s answers are reﬂected in the ﬁnal destination

recommendation.

In this study, therefore, a dialog system was developed

in which the robot takes the initiative in the dialog and, in

addition to the method of specifying the customer’s requests

with choice-type questions, a mechanism was introduced

to memorize the user’s speech and reﬂect the content of

the user’s speech in the recommendations. The objective

is to show through a dialog robot competition 2022 [4]

(DRC2022) what extent customers are satisﬁed with the sys-

tem and to what extent they are receptive to the destinations

recommended by the robot.

arXiv:2210.11223v1 [cs.RO] 20 Oct 2022

II. CISROBOT ARCHITECTURE

DRC2022 imposes a recommendation task on participants

of the competition, that is to recommend the one selected by

the operator in two sightseeing spots chosen by a participant.

Moreover, the limitation of time for the recommendation

task is for ﬁve minutes. Hence, a robot for DRC2022 has

to collect information about a participant and trip purposes

in a limited time. Therefore, we develop a framework for

managing a ﬂow of dialogues and equip into our robot named

CISRobot.

This section describes an architecture of our framework

named “Conversation Flow,” and does generalized functions

in our framework for enabling us to realize every dialogue

combined with the robot’s behaviors. Moreover, we also

describe common behaviors of CISRobot implemented to

increase user’s impression of our robot during conversations.

A. Conversation Flow

We aim to design CISRobot enabling us to certainly

perform the recommendation task. Hence, to complete the

task, a robot needs to collect information about a dialogue

partner and some purposes of their trip in a limited time. Our

framework “Conversation Flow (CF)” can manage a ﬂow

of conversations constructed by monologues and dialogues.

A robot speaks according to the contents of predeﬁned

monologues and dialogues so that the robot certainly listens

to information about a participant and the purposes of their

trip from users.

A CF is a series of three parts as follows:

1. Introduction:

In this part, a robot speaks ice break talks like

self-introduction and explanations about sightsee-

ing spots chosen by a participant.

2. a set of questions for collecting information:

After an ice-break talk in introduction part, a robot

begins collecting the required information for the

recommendation task. Hence, this part consists of

some questions for a participant and the robot’s

responses against its answer.

3. Conclusion:

A robot, in this part, concludes a conversation with

a participant, then a robot performs recommenda-

tion of sightseeing spots according to information

about participants collected in part 2.

Speciﬁcally, a robot that employed CF communicates with a

participant aligned with a ﬂow shown in Fig. 1. Part 2. has a

self-loop-shaped arrow; this arrow means to select questions

repeatedly as long as the left time for speaking conclusion

part. By modeling repeated selection at random, a robot can

speak different contents of conversation every time a robot

start-up. In another case, CF enables a robot to select ques-

tions preferentially depending on a given recommendation

task.

Fig. 2 partially shows a design of CF class, where

registIntrodcution(),registStartpoints(),

and registStartpoints() are member methods for

Fig. 1: A framework of our conversation ﬂow: We

design a ﬂow of conversations in this ﬁgure. Part 2. for

collecting participant’s information is a set of questions and

monologues deﬁned by a robot administrator in advance.

Fig. 2: This class diagram partially displays Conversation-

Flow class’s member variables and methods.

registering conversations for each part in CF, and

startConversation() is for beginning conversations

according with preregistered each part.

CF’s member variables introductionMonologue,

conclusionMonologue, and startpoints typed

“ConversationContent (CC)” class as shown in Fig. 3.

CC enables us to predeﬁne speciﬁcally a content

of conversations what a robot speaks or questions

is. Moreover, CC can register robot’s gestures by

invocating registBehaviorBeforseSpeech()

or registerBehaviorAfterSpeech() so that a

robot performs behaviors while speaking. Furthermore, CC

has two sub-classes as follows:

Monologue:

This sub-class enables us to deﬁne a content of

what a robot unilaterally speaks to a person is.

Hence, while a robot is speaking, microphones off.

Question:

This sub-class lets a robot question something to

a person, so it requires a sentence of a question.

At the same time, it needs to register two or

more CCs which are a robot’s line responding to a

person’s answer. When an instance of CC innovates

evaluate(), a CC selects a CC from registered

next conversations (nextConversations) by

evaluating whether there is a value of CC’s key

in an answer of a person. As we described above,

this implementation is rule-base.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

ADialogueRobotSystemtoImproveCredibilityinSightseeingSpotRecommendationsNaokiYoshimaru1,TomohiroMasuda2,HyejinHong2,YuseiTanaka2,MotoharuOkuma2,NagihiroMatsumoto2,KazumaKusu1,TakamasaIio2,andKenjiHatano2AbstractVariousstudieshavebeenconductedonhuman-supportingrobotsystems.Thesesystemshavebeenputtop...

展开>> 收起<<

A Dialogue Robot System to Improve Credibility in Sightseeing Spot Recommendations Naoki Yoshimaru1 Tomohiro Masuda2 Hyejin Hong2 Yusei Tanaka2 Motoharu Okuma2.pdf

共6页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

A Dialogue Robot System to Improve Credibility in Sightseeing Spot Recommendations Naoki Yoshimaru1 Tomohiro Masuda2 Hyejin Hong2 Yusei Tanaka2 Motoharu Okuma2

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: