FACILITATION SUPPORT FOR ON-LINE FOCUS GROUP

DISCUSSIONS BY MESSAGE FEATURE MAP

Noriko Imafuji Yasui

, Shunsuke Saruwatari

, Xavier Llor`a

and David E. Goldberg

IllGAL, University of Illinois at Urbana-Champaign, 104 S.Mathews Ave., Urbana, IL 61801, U.S.A.

NCSA, University of Illinois at Urbana-Champaign, 1205 W.Clark St., Urbana, IL 61801, U.S.A.

Keywords:

Focus group discussion, facilitation, message feature map.

Abstract:

Face-to-face focus group discussion has been one of the reliable approaches for collecting variety of ideas and

opinions for building marketing strategy, even though various network-based communication tools have been

available. This is due to complications in facilitation of on-line discussions. The goal of this paper is to maxi-

mize the proﬁt from on-line focus group discussions by supporting facilitators’ task. In this paper, we propose

a message feature map and two metrics for measuring message feature; centrality and novelty. The message

feature map plots discussion messages on centrality-novelty plane, and gives us intuitive understanding of the

discussions in various aspects. Reporting experimental results by using real data collected in on-line focus

group discussions, we discuss how we can utilize the message feature map for the effective facilitation.

1 INTRODUCTION

For these years, network-based communication has

come into wide use. People can easily access various

types of communication tools and use them for ob-

taining information, exchanging opinions and propa-

gating ideas. An objective of the focus group discus-

sions is to collect wide range of opinions and ideas.

Variety of ideas and opinions is highly valued. In case

of the on-line, we can obtain wide variety of many

ideas by having multiple discussions simultaneously

with various groups of diverse people.

The goal of this paper is to maximize the proﬁt

from on-line focus group discussions by supporting

discussion facilitators’ task. We propose a message

feature map, which is a visualization method for intu-

itive understanding of discussion status in various as-

pects. The message feature map plots each message

on the plane of the axes with the two metrics; central-

ity and novelty of each message. Centrality indicates

how much each message is center (or conversely, pe-

ripheral) in the discussion. Novelty tells us how much

each message contains novel (or conversely, conven-

tional) ideas.

2 MESSAGE FEATURE MAP

Success in on-line facilitation highly depends on how

quickly and easily they can analyze the discussions

in various aspects. This section proposes a message

feature map. After we introduce two metrics for mea-

suring message feature, we describe a classiﬁcation of

message characteristics into four types.

2.1 Measuring Message Feature

Centrality. This centrality metric measures how

much messages are center, conversely peripheral, in

the discussion. Suppose a discussion can be repre-

sented by a sequence of messages (m

, m

, ··· , m

By using the KEE algorithm ((Imafuji Y. et al.,

2007)), we calculate the score for each message k

times with different sets of messages. The message

centrality is obtained as the highest score in the k

message scores. Let M be a message score vec-

tor obtained by the KEE algorithm, and M(m

) be

the score for the message m

. Denote the central-

ity of the message m

by c(m

), c(m

) is deﬁned by

c(m

) = max

i≤ j<i+k

), (0 ≤ c(m

) < 1), where

) is a score of the message m

by the calculation

for a set of k messages {m

j− k+1

, ··· , m

j− 1

, m

Novelty. This novelty metric measures how much

messages include something new; ideas, topics, opin-

ions and etc. We assume that the message is novel,

if the messages contains a lot of terms which are not

previously used. Thus, we employ the simplest and

the easiest way for measuring novelty, that is, count-

563

Imafuji Yasui N., Saruwatari S., Llorà X. and E. Goldberg D. (2008).

FACILITATION SUPPORT FOR ON-LINE FOCUS GROUP DISCUSSIONS BY MESSAGE FEATURE MAP.

In Proceedings of the Tenth International Conference on Enterprise Information Systems - AIDSS, pages 563-566

DOI: 10.5220/0001705105630566

 SciTePress

center

peripheral

novel

conventional

Figure 1: Sample of the message feature map; plotting

messages on centrality (horizontal axis)- novelty (vertical

axis) plane. Four plotted areas are called np (upper-left), nc

(upper-right), cc (lower-right), and cp (lower-left).

ing the number of new terms in each message. We

deﬁne a new term in a message as a term which is not

appeared in some previous messages. If the novelty

of the message (or the message novelty, for short) is

high, the message would contain something new, and

initiate a new topic. Conversely, the low novelty in-

dicates the message would be following the existing

topic, or the message would be of no importance (in

case of the short messages).

Suppose a discussion can be represented by a se-

quence of messages (m

, m

, ··· , m

). Denote the

novelty of the message m

by n(m

), n(m

) is de-

ﬁned by n(m

) = N

), where N

) is a number

terms which are not existed in a set of the messages

i−l

, ··· , m

i−1

}. The novelty is also transformed so

that the values are in the range of -1 to 1.

2.2 Message Feature Classiﬁcation

The message feature map plots each message on

centrality-novelty plane. The Figure 1 is a sample

of a message feature map. The horizontal axis indi-

cates the centrality and the vertical axis indicates the

novelty. In this sample, messages are plotted with dif-

ferent colors (and shapes) for each discussant.

As seen in the ﬁgure, the message feature map is di-

vided into four areas; np (upper-left), nc (upper-right),

cc (lower-right), and cp (lower-left). The followings

are the plotted areas and the characteristic of the mes-

sages plotted in each area.

np -Potential chances. The messages plotted in this

area have high novelties and low centralities. The

messages of this type have new, but rare ideas, opin-

ions, topics and etc, which any other discussant is not

paying attention, nor talking about those topics any

further. These messages may turn out to be sources of

the ideas for new product or services in enterprises.

nc -Topic triggers. The messages plotted in this area

have high novelties and high centralities. The mes-

sages of this type lead new topics. The messages

bring new ideas or topics to the discussions. Originat-

ing from this message, the current discussion topics

are shifted to the topics of these messages. The leads

to discussion mainstream can be detected by observ-

ing these messages.

cc -Topic followers. The messages plotted in this area

have low novelties and high centralities. The mes-

sages of this type give more ideas or deeper insights

on the current topics. The discussion topics are some-

how converged by these messages. The discussion

mainstream can be detected by observing these mes-

sages.

cp -Triﬂes. The messages plotted in this area have

low novelties and low centralities. The messages of

this type do not inﬂuence on the discussion going.

The message content doesn’t have any speciﬁc topic.

Questions by discussion facilitators, yes and no an-

swers, and simply, short messages are plotted in this

area. The discussion is inactive if many messages are

plotted in this area.

3 EMPIRICAL STUDY

In the previous section, we proposed the message fea-

ture map, and described the message classiﬁcation on

the message feature map. This section studies empir-

ically how the message feature map supports facilita-

tors’ task. The discussion data (78 messages by six

participants) was collected from the focus group dis-

cussions held on March 2005. The goal of the discus-

sions was to identify ”future scenarios for cell phone

usages and the features that will make them popular

among consumers”. The data consists of a sequence

of messages (arranged in time order). A message con-

sists of message id, time, title, author name, replying

message id, and message content.

3.1 Effective Facilitation

Figure 2 depicts the message feature map using data

derived from one of the discussion group. The hor-

izontal axis and the vertical axis represent message

centrality and novelty, respectively. In this experi-

ment, we used k = 10 for the centrality, and l = 10

for the novelty. Each message is colored (or shaped)

differently for each participant.

The biggest advantage of the message feature map

is its comprehensibility. By looking at the map from

different angles, facilitators’ important work load, es-

pecially, analyzing and controlling, can be relieved.

Here are some effective use examples for the facilita-

tion support.

ICEIS 2008 - International Conference on Enterprise Information Systems

564

㪄㪈

㪄㪇㪅㪏

㪄㪇㪅㪍

㪄㪇㪅㪋

㪄㪇㪅㪉

㪇

㪇㪅㪉

㪇㪅㪋

㪇㪅㪍

㪇㪅㪏

㪈

㪄㪈㪄㪇㪅㪏㪄㪇㪅㪍㪄㪇㪅㪋㪄㪇㪅㪉㪇㪇㪅㪉㪇㪅㪋㪇㪅㪍㪇㪅㪏㪈

㪥㫆㫍㪼㫃㫋㫐

㪚㪼㫅㫋㫉㪸㫃㫀㫋㫐

㪧㪇㪈㪧㪇㪉㪧㪇㪊㪧㪇㪋㪧㪇㪌㪧㪇㪍

Figure 2: Message feature map derived from a real on-line focus group discussion. Each plot indicates a message, and each

plot shape represent each discussion participant.

Screening Messages. The more discussion groups

the facilitators have to take care, the more impractical

they read all the message contents. The message fea-

ture map tells intuitively which messages they should

focus on.

The messages plotted in the np area are contain-

ing many variable minority opinions and ideas. If the

facilitators want to ﬁnd rare, but creative ideas, they

should take a look at these messages. In the experi-

ment, there was a message ”one of the biggest things I

see for cell phones ... is money withdrawal. By linking

your cell phone and debit card, you could buy stuff

using cell phone and PIN number”. Although any

other participants never mentioned about debit cards

on phones, this idea might have potential into prac-

tice.

The messages plotted in the righter area contain

major topics that many other participants are talking

about. If the facilitators want to ﬁnd mainstream ideas

which are supported by many people, they should take

a look at these messages. In the experiment, one of the

messages plotted in nc area was about camera func-

tion on cell phone, which had not been popular in US

yet at the time of the discussion, and actually became

popular a while after the discussion.

Monitoring Participants. One of the biggest advan-

tages of the on-line focus group discussions is the par-

ticipants’ accessibility. We can have a series of dis-

cussions with various groups of participants by strate-

gically changing or grouping together some partic-

ipants. The message feature map gives us a quick

grasp behavior tendency of each participant.

For the example of Figure 2, four messages out

of eight in the nc area were by participant P01. We

could have an assumption that P01 was good at gen-

erating new topics. P02 wrote as many messages as

P01 wrote, but the messages by P01 are mostly plot-

ted in the cc area. This observation indicates that P01

has more potential of idea generation than P02 does.

The messages by P05 are not as many as P01 and P02,

but some of the messages are plotted in upper. P05

might display his creativity when his interests match

the discussion topics.

Controlling Discussion. One of the most critical

essences for success in discussions is to keep the dis-

cussions excited and the participants active. To do

so, the facilitators have to pay attentions to discus-

sion activeness and make sure that all the participants

actively enjoy the discussions. The message feature

map tells clearly when the discussion is getting stag-

nant.

Figure 3 shows the three message feature maps

(extracted from Figure 2); from 18th to 27th messages

(left), from 28th to 37th message (center) and from

38th to 47th (right). The messages in both left and

right maps are plotted over four areas, which mean

the discussion was very active during the periods. Es-

pecially, some messages in the right map are plot-

ted quite upper, which means the discussion was very

productive during the period. On the other hand, most

of the messages from 28th to 37th are plotted in the

area cc, which indicates that the discussion was con-

tracting and getting stagnant. If many messages come

to be plotted in the area cc, the facilitators have to

swing into action by asking something or talking to

the participants, whose messages are plotted in upper

area, for instance.

FACILITATION SUPPORT FOR ON-LINE FOCUS GROUP DISCUSSIONS BY MESSAGE FEATURE MAP

565

㪧㪇㪈㪧㪇㪉㪧㪇㪊㪧㪇㪋㪧㪇㪌㪧㪇㪍

㪄㪈

㪇

㪈

㪄㪈㪄㪇㪅㪌㪇㪇㪅㪌㪈

㪄㪈

㪇

㪈

㪄㪈㪄㪇㪅㪌㪇㪇㪅㪌㪈

㪄㪈

㪇

㪈

㪄㪈㪄㪇㪅㪌㪇㪇㪅㪌㪈

㪤㪼㫊㫊㪸㪾㪼㩷㪈㪏㪄㪉㪎

㪤㪼㫊㫊㪸㪾㪼㩷㪉㪏㪄㪊㪎

㪤㪼㫊㫊㪸㪾㪼㩷㪊㪏㪄㪋㪎

Figure 3: Message feature maps for the messages 18th-27th (left), 28th-37th (middle), and 38th-47th (right).

4 CONCLUSIONS

In this paper, we proposed a message feature map,

which was a visualization for plotting discussion mes-

sages on a plane with two axes. We also introduced

two metrics; centrality and novelty. This map gave us

an intuitive understanding and quick grasp of discus-

sion status. We showed the experimental results with

using the data collected in a real on-line focus group

discussion, and presented some scenarios for facilita-

tion support usages by the message feature maps.

Our future works include to build a message tran-

sition model which tells how message status transit on

the message feature map, and to simulate discussion

for a given set of participants. Discussion simulation

will be a very useful tool for discussion planning -

determining the discussion goal, grouping the people,

building strategic facilitation scenarios.

5 RELATED WORKS

Several tools for on-line focus group

discussions have been introduced for

these years (Zoomerang Online Focus

http://info.zoomerang.com/prodserv/onlinefocus.htm,

GMI Focus Group Software https://www.gmi-

mr.com/net-mr/online-focus-groups.php, and e-

Focusgroups http://www.efocusgroups.co.uk/). For

example, MarketTools, Inc. has launched Zoomerang

Online Focus, which is a web-based solution that

helps marketers conduct focus group research on-

line. One of its selling points is that it provides

highly skilled and trained facilitators (, which they

call moderators). Our work aims to have proﬁtable

on-line focus group discussions without depending

on facilitators’ skills.

Various methods have been proposed for ﬁnding

important terms from text (key phrases (Witten et al.,

1999), topic words (Lawrie et al., 2001)). Some

works have focused on ﬁnding key persons in text-

based communication (Kamimaeda et al., 2005; Re-

ich et al., 2002), and on exploring social networks

of network-based communication (Zhou et al., 2006).

These are very effective and high quality analyses,

which give us deep insight into each aspect. In or-

der to analyze in various aspects, they need to use

multiple analyses and compare them. This might be

time-consuming work for facilitators. Our goal is to

propose methods or tools which give us multi infor-

mative analysis on a single output.

ACKNOWLEDGEMENTS

We would like to thank to Hakuhodo Inc. for their

project collaboration. This work was sponsored by

the Air Force Ofﬁce of Scientiﬁc Research, Air Force

Materiel Command, USAF (AF9550-06-1-0096 and

AF9550-06-1-0370).

REFERENCES

Imafuji Y., N., Llor`a, X., Goldberg, D. E., Y., W., and D., T.

(2007). Delineating topic and discussant transitions in

online collaborative environments. In Proceedings of

9th International Conference on Enterprise Informa-

tion Systems (ICEIS 2007).

Kamimaeda, N., Izumi, N., and Hasida, K. (2005). Dis-

covery of key persons in knowledge creation based on

semantic authoring. In KMAP 2005.

Lawrie, D., Croft, W. B., and Rosenberg, A. (2001). Finding

topic words for hierarchical summarization. In SIGIR

’01: the 24th ACM SIGIR conference on Research and

development in information retrieval, pages 349–357.

Reich, J. R., Brockhausen, P., Lau, T., and Reimer, U.

(2002). Ontology-based skills management: Goals,

opportunities and challenges. Universal Computer

Science, 8(5):506–515.

Witten, I. H., Paynter, G. W., Frank, E., Gutwin, C., and

Nevill-Manning, C. G. (1999). Kea: practical auto-

matic keyphrase extraction. In DL ’99: the fourth

ACM conference on Digital libraries, pages 254–255.

Zhou, D., Manavoglu, E., Li, J., Gies, C., and Zha,

H. (2006). Probabilistic models for discovering e-

communities. In Proceedings of the 15th international

conference on World Wide Web (WWW2006).

ICEIS 2008 - International Conference on Enterprise Information Systems

566