Experimental Investigation for a Human Relationship Formation

Support Agent using Information Presentation During Conversation

Michimasa Inaba, Kana Otsuka and Kenichi Takahashi

Graduate School of Information Sciences, Hiroshima City University,

3-4-1 Ozukahigashi, Asaminami-ku, Hiroshima, Japan

Keywords:

Communication Support Agent, Information Presentation, Intelligent User Interface.

Abstract:

In this paper, we performed an experimental investigation aimed at developing an agent to support the for-

mation of human relationships by supporting the user’s daily communication “casually”, “anytime” and “any-

where”. First, we collected conversations between men and women meeting for the ﬁrst time, then analyzed

what type of support would be effective for the formation of human relationships. Based on the results of this

analysis, we performed experiments supporting communication. In the experiment, we provided not only top-

ics to the user during conversation, but also comprehensive presentation of instructions such as expressions,

eye contact and gestures. The results conﬁrmed that this signiﬁcantly improved human relationships after

conversation and showed the validity of this support.

1 INTRODUCTION

Our goal is to realize an agent that will grasp the com-

munication state of a supported user and his or her

conversation partner in real-time and offer advice to

the user through a wearable device. Studies to support

communication have been performed widely. How-

ever, these have been extremely limited in terms of

use conditions, requiring large equipment or displays

or for the conversation participants to wear special-

ized devices such as sensors. The objective of this

study is to support the formation of human relation-

ships by supporting the user’s daily communication

“casually”, “anytime” and “anywhere”.

This study does not suppose any specialized situa-

tions, such as the conversation partner using a partic-

ular device or being equipped with sensors. Only the

supported user wears the required device, and support

is given independent of situation or location. More-

over, the system provides not only topics to the user

during conversation, but also includes comprehensive

presentation of instructions such as expressions, eye

contact and gestures. We aim to realize an agent that

makes it possible to promote better human relation-

ship formation through those presentations.

However, it is not necessarily clear what types of

content would be effective to present to the user, or

that it would be possible for such a system to promote

the formation of good human relationships. Thus, in

this study we collect data from conversations between

men and women meeting for the ﬁrst time and per-

form analysis to ﬁnd effective support content. Then,

we perform experiments that present the support con-

tent to the user during communication. In this study,

however, we do not target situational recognition dur-

ing communication, but have a person to choose the

content and timing to be presented in the experiment.

We then assess the content presented to the user and

the effectiveness of support during communication

from the perspective of forming better human rela-

tionships. Also, in this study we viewed mutual in-

terpersonal impressions as a human relationship, and

examined the effectiveness of promoting human re-

lationship formation by determining how much in-

terpersonal impressions of conversations partners im-

proved during conversation.

Participants answered a survey covering a number

of impressions before and after the conversation, and

data was collected on the changes. Next, we analyzed

how the conversations related to changes in impres-

sions. Through this analysis, we determined the con-

tent to present. Finally, we performed an experiment

to support communication by presenting the content

determined through the analysis to a user. However,

we do not target situational recognition during com-

munication, rather having a person to choose the con-

tent and timing to be presented.

Inaba, M., Otsuka, K. and Takahashi, K.

Experimental Investigation for a Human Relationship Formation Support Agent using Information Presentation During Conversation.

DOI: 10.5220/0005643200870094

In Proceedings of the 8th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2016) - Volume 1, pages 87-94

ISBN: 978-989-758-172-4

2 RELATED RESEARCH

A number of studies that present information to con-

versation participants in order to support face-to-face

communication are being carried out.

Research on supporting communication through

the presenting information on a display or other de-

vice includes the Ambient Suite information environ-

ment proposed by Fujita et al. to support communi-

cation between multiple people (Fujita et al., 2011).

Ambient Suite is a system that estimates the state of

the conversation using multiple sensor attached to the

participants and displays photographs or other infor-

mation on displays placed on the walls or ﬂoor. Multi-

ple studies to support communication in meetings and

similar environments by visualizing elements such as

talk time, gaze and head movement are also being per-

formed (Madan et al., 2004; Sturm et al., 2007). Kim

et al. proposed Meeting Mediator, which supports

balanced discussions by visualizing participant talk

amounts on portable terminals during meetings (Kim

et al., 2008). Ticket2Talk, proposed by Mccarthy et

al., is a system that displays the area of expertise of

conference participants standing near a display using

RFID tags(McCarthy et al., 2004). Other studies in-

clude a method to present information related to con-

versation participants by attaching miniature displays,

such as badges (Borovoy et al., 1998) and a method to

present information in informal spaces such as break

rooms (Matsuda and Nishimoto, 2002).

However, the systems in these studies place ma-

jor restriction on the place and situation in which they

can be used, such as requiring large devices or par-

ticipants wearing sensors or RFID tags. In addition,

support effectiveness is assessed through increases in

conversation (Fujita et al., 2011; Matsuda and Nishi-

moto, 2002) or satisfaction (Kim et al., 2008; Mc-

Carthy et al., 2004). We are not aware of any studies

which treat promoting the formation of human rela-

tionships as the main objective and perform analysis

from that aspect.

3 COLLECTION AND

ANNOTATION OF

CONVERSATION DATA

In this study, we ﬁrst collected conversation data for

determining the instruction content to be presented

to the user. The collected data was taken at ﬁfteen-

minute intervals from a male and female meeting for

the ﬁrst time. Questionnaire surveys were conducted

immediately before and after the conversation.

In this chapter, we talk about the conversation data

collection method and questionnaire content.

3.1 Collection Method

In this study, we collected data from conversations

between unacquainted males and females meeting for

the ﬁrst time. This type of conversation was chosen in

order to make it easier to analyze the relationship with

the actual communication carried out by starting with

a situation in which human relationships were not yet

formed, then observing what type of relationship was

formed after the the conversation ended. We also con-

sidered that using conversations between males and

females would more easily show differences in re-

gards to conversation liveliness and participant atti-

tude in comparison with conversations between two

members of the same gender.

The conversation collection procedure is as fol-

lows. First, participants were moved from waiting

rooms, which were divided by gender, to a room for

collection. The participants sat across from one an-

other at a table. Upon a signal from an operator, par-

ticipants greeted one another and gave self introduc-

tions. Afterwards, the participants returned to their

waiting rooms and answered a pre-conversation sur-

vey. After answering the survey, both participants

returned to the collection room and sat down once

again. At a signal from the operator,they began a con-

versation lasting for ﬁfteen minutes. No restrictions,

instructions or proposals were given for the topics or

content discussed by the participants; they were only

told to converse for ﬁfteen minutes. After the conver-

sation, each participant moved to another room where

he or she answered a post-conversation survey.

3.2 Survey

In order to grasp changes in the relationship between

the participants before and after the conversation, par-

ticipant personalities and characteristics, and conver-

sation conditions, we gave surveys before and after

the ﬁfteen-minute conversation. The content of these

surveys are described below.

3.2.1 Pre-conversation Survey

The pre-conversation survey was composed of two

items using an eight-step Lickert scale (1: very good,

8: not good at all). Survey items are shown below.

[Eight-step Survey Items]

Q1. Do you have a good impression of your conver-

sation partner?

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

Q2. Are you good at small talk?

The ﬁrst eight-step item, Q1, is a question to sur-

vey the participant’s impression of his or her partner.

The second, Q2, is a question to survey the partici-

pant’s conversation ability.

3.2.2 Post-conversation Survey

The post-conversation survey was composed of

three items: two eight-step Lickert scale items (1:

very good, 8: not good at all) and one free writing

question. The survey items are shown below.

[Eight-step Survey Items]

Q1. Do you have a good current impression of your

conversation partner?

Q2. Did the conversation go well?

[Free Writing Item]

Q3. Please write down anything that you noticed in

particular.

The ﬁrst eight-step item, Q1, is a question to ex-

amine the participant’s impression of his or her part-

ner. The second, Q2, is a question to survey his or her

impression of the conversation itself. The free writing

item, Q3, is a question to survey anything the partici-

pant noticed during the conversation.

3.3 Collection Results

The participants consisted of a total of 26 college stu-

dents, graduate students and high school students (13

males: average age 21.92, 13 females: average age

21.46, overall: average age 21:69). A total of 23 con-

versations were carried out.

Table 1 shows the results of the pre-conversation

survey conducted together with conversation data col-

lection, and Table 2 shows the results of the post-

conversation survey. Comparing the average values

for the question “Do you have a good impression of

your conversation partner?” before the conversation

and “Do you have a good current impression of your

conversation partner?” after the conversation, we ﬁnd

that the value increased by 0.72. Thus, we ﬁnd that

conversations tended to improve impressions.

Looking at the question “Are you good at small

talk?”, which examines the participant’s own conver-

sation ability, we see that the average value of 4.58 fell

roughly in the center of the eight-step scale. The vari-

ance was also larger in comparison with other items,

showing that the participants included both those who

Table 1: Pre-conversation survey results (conversation data

collection).

Survey item Average Variance

Q1 Do you have a good 2.59 1.05

impression of your

conversation partner?

Q2 Are you good at 4.58 3.69

small talk?

Table 2: Post-conversation survey results (conversation data

collection).

Survey item Average Variance

Q1 Do you have a good 1.87 0.52

current impression of

your conversation

partner?

Q2 Did the conversation 2.57 1.67

go well?

were skilled at conversation and those who were not.

In the “Did the conversation go well?” question in

the post-conversation survey, we ﬁnd a positive result

with an average of 2.57. This suggests that conversa-

tions tended to go well overall.

3.4 Instruction Content

Based on analysis of the collected conversation data

and our literature review, we determined content to be

presented to a target user for communication support.

Table 3 shows the instruction content we determined.

The line “(2 from Topics A)” seen in 3, 5, 6, 7 and

9 of the table indicate choosing and displaying the

speciﬁed number of topics from those given in Table

4. For example, the instruction content of 3 in Table

3 proposes “Ask your partner about his or her plans

“Christmas” “Travel” ”.

Below, we given our reasons for determining each

instruction content. Instruction contents 1 to 9 were

determined based on the determination made from

the results of the analysis of the collected conversa-

tion data, i.e. that if the participants who become

the speaker and listener, respectively, actively pursue

their own respective roles, this tends to improve im-

pressions. In addition, instruction contents 1 to 4 are

intended to support a listener in moving the conver-

sation forward, while 5 to 9 are intended to support a

speaker. The topics in Table 4 were determined with

reference to the list of topics for initial conversations

between college students given by Mimaki (Mimaki,

2013).

Instruction content 10 is to be given in situations

in a conversation when the participant can nod, based

Experimental Investigation for a Human Relationship Formation Support Agent using Information Presentation During Conversation

Table 3: Instruction content.

No. Content

1 Ask your partner “What about you?” or

“What do you think?”

2 Ask for a reason

3 Ask your partner about his or her plans (2

from Topics A)

4 Ask your partner about him/herself

“Hometown”, “Hobbies, recent interests”,

“Siblings”

5 Let your partner speak about the following

topic (3 from Topics B)

6 Talk about yourself (2 from Topics C)

7 Talk about your own plans (2 from Topics

8 Talk about your own experiences “Part-

time job”, “Travel”, “Club activities”

9 Talk about one of the following topics con-

necting with yourself (3 from Topics B)

10 Nod in agreement with your partner

11 Respond with “I see” or “Yes, that’s true”

12 Do not cover your hand with your face

13 Do not put your elbows up

14 Do not touch your hair

15 Do not cross your arms

16 Try to smile

17 Talk about the following topics to ﬁnd

common points with your partner “Hob-

bies”, “Sports”, “Favorite music”, “Home-

town”

18 Look your partner in the eye and speak

19 Try to maintain good posture

20 Try to maintain a bright tone of voice

Table 4: Topics to present.

Topics A Christmas, Travel, New Year’s

Eve, New Year’s Day, End-of-year

cleaning

Topics B TV, Pro baseball, School, Movies,

J-League soccer, Studies, Video

games, Food, Sports

Topics C Hometown, Hobbies/current inter-

ests, siblings, recent events

on the analysis results, in which it was determined

that more nodding tended to improve impressions. In-

struction content 11 is also meant to indicate to the

partner that one is listening, as with nodding. Instruc-

tion contents 12 to 15 were determined from the anal-

ysis results, which determined that touching hair or

crossing arms can harm one’s impression. Instruction

content 16 is given based on the analysis, in which

it was determined that smiling tended to improve im-

pressions.

Instruction content 17 was determined to make it

easier to discover common points with one’s partner,

as the free writing section of the post-conversation

survey included multiple opinions such as “the con-

versation went well because we were able to ﬁnd

common points”, and in these cases impressions im-

proved.

Instruction contents 18 to 20 are given to give a

better impression to one’s partner.

4 CONVERSATION SUPPORT

THROUGH INFORMATION

INSTRUCTIONS

In this chapter, we perform an experiment in which

we support a conversationby teaching the participants

information. We then compare these with conversa-

tions without instructions to determine whether they

were able to form better human relationships. Follow-

ing the procedure of the data collection experiment

conducted in Chapter 3, males and females meeting

for the ﬁrst time conducted ﬁfteen-minute conver-

sations with pre-conversation and post-conversation

surveys.

During the experiment, one of the two participants

received information. The information was presented

using a projector and screen. This was done because

we believed it would be best to avoid wearing any

special apparatus and to simulate the conditions of

the conversation collection experiment in Chapter 3

as closely as possible. The content and timing of the

content shown on the screen was determined manu-

ally. In this experiment, an operator determines the

state of the conversation and decides the content to

display.

4.1 Instruction Method

Instructions were presented using a projector and

screen. While sounds could be used to convey in-

structions in addition to visual cues, this could make

it more difﬁcult to concentrate on the conversation, so

in this experiment we gave instructions using a pro-

jector and screen.

Figure 1 shows an outline of the experiment envi-

ronment. Of the two participants shown in the ﬁgure,

one is the instructed participant, who will receivecon-

versation support in the form of instructional informa-

tion, and the other is the non-instructed participant,

who will not receive support. The two participants sit

across a table from one another and conduct a conver-

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

Figure 1: Experiment environment orientation.

Figure 2: Screen display examples (translated by authors).

sation. A project and screen are set up behind the non-

instructed participant. In this manner, instructions are

presented only to the instructed participant. The op-

erator is seated in a position that allows him or her to

see the expressions and movements of the instructed

participant.

The operator observes the conversation content as

well as the actions and attitude of the participants and

displays instructions suitable to the conditions on the

screen by operating a PC connected to the projec-

tor. Figure 2 gives four examples of information dis-

played to the screen (Original instructions are written

in Japanese only). Text is displayed on the top half of

the screen to make it easier to see over the conversa-

tion partner.

When the conversation begins, topics for ﬁnding

common points are presented (17 in Table 3). From

then on, the operator determines the content and tim-

ing of the displayed information based on the follow-

ing basic guideline.

• Provide topics during periods of silence (4, 6, etc.

in Table 3)

• Point out issues with the participant (12, 13, etc.

in Table 3)

• Promote responses (10, 11, etc. in Table3)

• In cases when there is little need for support, such

as when the conversation is going well, display

16, 18, 19, 20 in Table 3 in order.

Instruction content does not disappear from the

screen until the next instruction content is displayed.

In other words, the screen is always displaying some

information until the conversation ends. However, the

operator gavenew content for display at least once per

minute. The same instruction was not given twice in

a row.

As a result of the analysis, we determined that lis-

teners and speakers should concentrate on their re-

spective roles. Thus, we determined whether the

instructed participant was a listener or a speaker

and changed instruction content accordingly. As de-

scribed in the previous section, 1 to 4 in Table 3 cor-

respond with listener instructions, while 5 to 9 corre-

spond with speaker instructions. The determination of

whether the participant was a speaker or a listener was

made by the operator during roughly the ﬁrst minute

of the conversation.

4.2 Experiment Method

The experiment was conducted with male and female

participants meeting for the ﬁrst time. Participants

were composed of a total of 20 college and graduate

students (10 males: average age 22.10, 10 females:

average age 22.00, overall: average age 22.05). A

total of 20 conversations were conducted. Each par-

ticipant conducted conversations with two members

of the opposite gender. In one of these conversations

the participant received instruction, while in the other

he or she did not receive instruction (while his or her

partner received information).

The experiment procedure followed that of the

conversation data collection experiment described in

Section 3.1. In other words, participants greeted one

another, performed self introductions, completed pre-

conversation surveys, conducted a ﬁfteen-minute con-

versation, and completed post-conversationsurveys in

that order. However, in this experiment we explained

the instruction method and content to both the in-

structed participant and non-instructed participant be-

fore greetings and self-introductions. This informa-

tion was also explained to the non-instructed partici-

pant in order to preventconditions in which the partic-

ipant became concerned with what was displayed on

the screen and could not concentrate on the conversa-

tion. Instructed participants were told that while they

Experimental Investigation for a Human Relationship Formation Support Agent using Information Presentation During Conversation

did not have to follow all instructions if the actions

or conversation felt unnatural, they should follow the

instructions as closely as possible.

4.3 Surveys

The pre-conversation survey was identical to that

given in the conversation data collection experiment

in Section 3.2.1. Both instructed participants and non-

instructed participants used the same pre-conversation

survey, but each answered different surveys after the

experiment. In addition to the post-conversation sur-

vey explained below, the instructed participant also

completed a survey examining the utility of the in-

struction content. This was performed by giving the

instructed participant a list of the instruction content

shown on the screen and having him or her mark

which items were useful and which were not (with

multiple answers acceptable).

4.3.1 Post-conversation Survey for Instructed

Participants

The post-conversation survey for instructed partic-

ipants consisted of eight eight-step Lickert scale

items (e.g. 1: strongly agree, 8: strongly disagree),

one ten-step Lickert scale item (1: 10%, 10: 100%),

and two free writing items. Survey items are shown

below.

[Eight-step Survey Items]

Q1. Do you have a good current impression of your

conversation partner?

Q2. Did the conversation go well?

Q3. Was the instruction timing good?

Q4. Were the text position, size and color good?

Q5. Were the instructions useful for smoothly mov-

ing the conversation forward?

Q6. Were the instructions useful for making the con-

versation go well?

Q7. Were the pointers and advice for attitude and be-

havior (e.g. “Try to smile” or “Nod”) useful?

Q8. Overall, were the instructions useful?

[Ten-step Survey Item]

Q9. To what extent were you able to utilize the

conversation support contents?

[Free Writing Items]

Q10. Please writing any positive points about the in-

structions

Q11. Please writing any points about the instructions

that need improvement

Of these, Q1 and Q2 are the same as those used

in conversation data collection. Q3 to Q8 are used

to assess the instructions from the instructed partici-

pant’s point of view. The single ten-step item is used

to survey how much the participant followed the in-

structions. Free writing items Q10 and Q11 are used

to survey opinions related to instructions given during

conversation.

4.3.2 Post-conversation Survey for

Non-instructed Participants

The survey for non-instructed participants was

composed of three eight-step Lickert scale items. The

survey items are shown below.

[Eight-step Survey Items]

Q1. Do you have a good current impression of your

conversation partner?

Q2. Do you think the conversation went well?

Q3. Did you feel any unease about your conversation

partner being supported?

Of these, Q1 and Q2 are the same as those used

in the conversation data collection experiment. Q3 is

used to assess the instructions from the non-instructed

participant’s point of view.

4.4 Experiment Results and Discussion

Table 5 shows the result of the pre-conversation sur-

vey, Table 6 shows the results of the instructed par-

ticipant post-conversation survey, and Table 7 shows

the results of the non-instructed participant post-

conversation survey. Instructions were shown an av-

erage of 15.1 times during conversation.

In Q2 “Are you good at small talk?” of Table 5, the

average value is 4.50. As with the conversation data

collection experiment, participants both good and bad

at small talk participated. In Q9 of Table 6, the uti-

lization of the instruction contents was about 60

Q3 in Table 7 shows that non-instructed partici-

pants felt little discomfort about their partners receiv-

ing instructions. This suggests that the visual presen-

tation method was suitable.

4.4.1 Changes in Impression Due to the

Presence of Instructions

As with the data analysis, we examined the change

in partner impression by looking at the variance be-

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

Table 5: Pre-conversation survey results (conversation sup-

port experiment).

Survey item Ratio Variance

Q1 Do you have a good

impression of your

conversation partner?

3.00 1.33

Q2 Are you good at

small talk?

4.50 3.74

Table 6: Results of post-conversation survey for instructed

participants (conversation support experiment).

Survey item Ratio Variance

Q1 Do you have a good

current impression

of your conversation

partner?

2.10 1.57

Q2 Do you think the con-

versation went well?

3.05 2.79

Q3 Was the instruction

timing good?

3.30 1.59

Q4 Were the text position,

size and color good?

2.30 1.38

Q5 Were the instructions

useful for smoothly

moving the conversa-

tion forward?

2.95 1.84

Q6 Were the instructions

useful for making the

conversation go well?

3.20 2.59

Q7 Were the pointers and

advice for attitude and

behavior (e.g. “Try to

smile” or “Nod”) use-

ful?

2.95 1.21

Q8 Overall, were the in-

structions useful?

3.05 1.63

Q9 To what extent were

you able to utilize

the conversation sup-

port contents?

6.30 6.64

tween the answer to the eight-step item Q1 in the pre-

conversation survey and the answer to the eight-step

item Q1 in the post-conversation survey. As seen in

Tables 5 through 7, the average changes in impression

for participants were 0.70 for non-instructed partici-

pants and 1.15 for instructed participants. The aver-

age value for changes in impression in the data col-

lection experiment in Chapter 3 was 0.72

These results show that the impression of in-

structed participants improved greatly in comparison

Participants in the conversation data collection experi-

ment can also be viewed as non-instructed participants.

Table 7: Results of post-conversation survey for non-

instructed participants (conversation support experiment).

Survey item Ratio Variance

Q1 Do you have a good

current impression

of your conversation

partner?

2.05 0.37

Q2 Do you think the con-

versation went well?

2.35 0.87

Q3 Did you feel any un-

ease about your con-

versation partner being

supported?

5.85 2.13

with participants in the conversation data collection

experiment and non-instructed participants in this ex-

periment. In addition, a t-test conﬁrmed a signif-

icant difference with both with a signiﬁcance level

of 5%. Thus, this shows that the impressions of in-

structed participants improve signiﬁcantly in compar-

ison with non-instructed participants. Thus, we were

able to conﬁrm that conversation support through in-

structions is useful for enlivening conversation and

obtaining better impressions from others.

4.4.2 Effectiveness of Conversation Support

Through Information Presentation

Survey results for Q5 to Q8 of Table 6 give assess-

ments of approximately 3.0 for instructions helping

to smoothly move conversations forward, enlivening

conversations and helping to face conversations with

better attitudes and behaviors. We obtained good re-

sults for other subjective assessments from the in-

structed participants as well.

In addition, in analyzing the free writing survey,

we found opinions from participants that answered

that the instructions were useful included that topic

support “was good because I was given topics when

I couldn’t think of any”, and that “the conversa-

tion started smoothly because the ﬁrst topic was pro-

vided”. There were also opinions about the indica-

tions and advice for attitude and behavior, includ-

ing “I was nervous about talking with someone I was

meeting for the ﬁrst time, but the indications about

behavior helped me to smile”.

4.4.3 Instruction Content Assessment

Table 8 shows the results of the survey related to the

usefulness of instruction content conducted simulta-

neously with the post-conversation survey. The ta-

ble shows ratios of participants who answered that

instructions were helpful or not helpful for seven in-

Experimental Investigation for a Human Relationship Formation Support Agent using Information Presentation During Conversation

Table 8: Instruction content usefulness assessment results.

Instruction content Ratio of useful deter-

minations

Ratio of non-useful

determinations

Ask your partner about him/herself “Hometown”,

“Hobbies, recent interests”, “Siblings”

0.90 0.00

Look your partner in the eye and speak 0.80 0.00

Try to maintain good posture 0.70 0.05

Try to smile 0.65 0.10

Nod in agreement with your partner 0.60 0.05

Respond with “I see” or “Yes, that’s true” 0.50 0.00

Try to maintain a bright tone of voice 0.40 0.15

struction content items that were shown during ten or

more (half of the total) conversations.

In the table, six items were determined to be useful

by a ratio of more than 0.50 of participants. Ratios of

participants answering that these same six items were

not useful were less than 0.10.

On the other hand, the instruction to brighten

one’s town of voice had the lowest ratio of partici-

pants who found it useful at 0.50, and the highest ratio

of participants who did not ﬁnd it to be useful at 0.15,

the highest of the seven items. It may be that because

suddenly raising one’s voice would be unnatural, and

many participants found this difﬁcult to utilize, mak-

ing it less useful. Thus, we must examine changing

expressions or giving more speciﬁc instructions for

raising the tone of one’s voice.

5 CONCLUSION

In this study, we aimed to construct a system that

would allow promoting the formation of human re-

lationships through displaying information to a user,

and showed instruction content and the usefulness of

conversation support through information presenta-

tion experimentally.

We ﬁrst collected conversation data between two

people meeting for the ﬁrst time.Next, we analyzed

the conversation and using the knowledge gained

from the results of the analysis, we determined con-

tent that should be presented to the user in order to

give his or her partner a good impression. Finally,

upon conducting an experiment supporting conver-

sation through presenting information to a user dur-

ing conversation, we conﬁrmed that these instructions

signiﬁcantly improved the impression of the user’s

partner compared with users that did not receive in-

struction.

REFERENCES

Borovoy, R., Martin, F., Vemuri, S., Resnick, M., Silver-

man, B., and Hancock, C. (1998). Meme tags and

community mirrors: moving from conferences to col-

laboration. In Proceedings of the 1998 ACM confer-

ence on Computer supported cooperative work, pages

159–168. ACM.

Fujita, K., Itoh, Y., Ohsaki, H., Ono, N., Kagawa, K.,

Takashima, K., Tsugawa, S., Nakajima, K., Hayashi,

Y., and Kishino, F. (2011). Ambient suite: enhancing

communication among multiple participants. In Pro-

ceedings of the 8th International Conference on Ad-

vances in Computer Entertainment Technology, pages

25:1–25:8. ACM.

Kim, T., Chang, A., Holland, L., and Pentland, A. S. (2008).

Meeting mediator: enhancing group collaborationus-

ing sociometric feedback. In Proceedings of the 2008

ACM conference on Computer supported cooperative

work, pages 457–466. ACM.

Madan, A., Caneel, R., and Pentland, A. S. (2004). Group-

media: distributed multi-modal interfaces. In Pro-

ceedings of the 6th international conference on Multi-

modal interfaces, pages 309–316. ACM.

Matsuda, K. and Nishimoto, K. (2002). Huneas: Supporting

informationsharing and activating human-network by

exploiting spontaneous encounters in an organization.

IPSJ Journal, 43(12):3571–3581.

McCarthy, J. F., McDonald, D. W., Soroczak, S., Nguyen,

D. H., and Rashid, A. M. (2004). Augmenting the so-

cial space of an academic conference. In Proceedings

of the 2004 ACM conference on Computer supported

cooperative work, pages 39–48. ACM.

Mimaki, Y. (2013). Discourse Analysis on Politeness : How

People Interact at First Encounters. Kuroshio Pub-

lishers.

Sturm, J., Herwijnen, O. H.-v., Eyck, A., and Terken,

J. (2007). Inﬂuencing social dynamics in meetings

through a peripheral display. In Proceedings of the 9th

international conference on Multimodal interfaces,

pages 263–270. ACM.

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence