ditlab system for Dialogue Robot Competition 2022 Yuuki Tachioka1 Abstract We developed a dialogue system for Dialogue

2025-04-26 0 0 358.02KB 3 页 10玖币

侵权投诉

ditlab system for Dialogue Robot Competition 2022

Yuuki Tachioka1

Abstract— We developed a dialogue system for Dialogue

Robot Competition 2022. Our system is composed of three

parts. First part investigates participants’ demographic infor-

mation by rule-based interview. Second part recommends a

point of interest (POI) based on the collected demographic

information. Third part answers participants’ question based

on the combination of rule-based answering and deep-learning-

based answering with nearby POI search.

I. INTRODUCTION

Dialogue Robot Competition 2022 is a travel agency

dialogue task, which aims to develop a dialogue service with

hospitality. Detailed conditions are described in the papers

published by the organizers [1], [2]. Our system honors

customers’ preference and assists customers’ decision. Types

of point of interest (POI) are classiﬁed into sightseeing type

or experience type. Interview with customers obtains demo-

graphic information and determines which POI is preferable

for the customer. System recommends a preferable POI and

explains a ground based on the demographic attributions

or travel conditions. System answers customers’ question in

two types of systems that pick up a corresponding pair of

question and answer based on a keyword search or generate

answers by a neural-based system. In addition, to make the

given information more attractive, nearby POIs are searched

and POI with better reputation is recommended.

II. DIALOGUE FLOW

A. Overview of system

Fig. 1 shows an overview of the system composed of

two elements. First element makes a recommendation of

POI considering customers’ attribution of demographic infor-

mation or preference estimated from customers’ interviews.

To estimate customer’s preference, system uses collected

information or emotion recognition results. Second element

is a question and answer part. Two types of methods are

used to make answers: rule-based one or neural dialogue

generation. Instead of ﬁnding a corresponding question and

answer from entire question and answer database, system

collates a question corresponding to the target category after

category estimation of a customers’ question. If appropriate

answer cannot be found in question and answer pairs, neural

dialogue generation can generate an answer. In addition,

nearby POI is searched, if a customer is interested.

*This system had been constructed for the Dialogue Robot Competition

2021 with Atsushi Keyaki who worked at Denso IT Laboraory

1Denso IT Laboratory, 2-15-1 Shibuya, Shibuya, Tokyo, Japan

tachioka.yuki@core.d-itlab.co.jp

Fig. 1. Overview of our system

B. Investigation of demographic information

Demographic information is extracted from an interview

with a customer. First, our system asks a name of customers

to make the customer feel familiar with the system. To avoid

misunderstanding, if answer is recognized as famous family

names (top 5,000 in Japan), the recognition result is adopted

and this is used for calling customers. Second, to clarify

the customer’s demographic information and preference,

following questions are asked to the customers.

1) How many times did you visit Odaiba?

2) How many people are you accompanying with?

3) Which types of travel is favorite? (sightseeing type:

watching exhibition as you like; experience type: ex-

perience something by yourself)

4) (In the case of experience type and if the number

of accompanying person is 3 and more,) do you

accompany small children?

5) (If recommended POIs can allow visitors to accompany

pet,) do you intend to accompany pets?

C. Recommendation

Recommendation is made by grounds. Fig. 2 shows an

example of recommendation. System ﬁnds grounds of rec-

ommendation from demographic attributions such as age, ac-

companying persons, and preferences. There are two types of

POIs: sightseeing type or experience type. User’s preferences

are matched with a type of POI and the matched POI is

recommended with grounds. Information of POI is collected

by web search (Jalan1or Google Map)

1https://www.jalan.net/kankou/

arXiv:2210.06646v1 [cs.RO] 13 Oct 2022

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：

ditlabsystemforDialogueRobotCompetition2022YuukiTachioka1AbstractWedevelopedadialoguesystemforDialogueRobotCompetition2022.Oursystemiscomposedofthreeparts.Firstpartinvestigatesparticipants'demographicinfor-mationbyrule-basedinterview.Secondpartrecommendsapointofinterest(POI)basedonthecollecteddemog...

展开>> 收起<<

ditlab system for Dialogue Robot Competition 2022 Yuuki Tachioka1 Abstract We developed a dialogue system for Dialogue.pdf

共3页,预览1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！

ditlab system for Dialogue Robot Competition 2022 Yuuki Tachioka1 Abstract We developed a dialogue system for Dialogue

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: