■ We applied mobile computing to community sup-port and explored mobile computing with a largenumber of terminals. This article reports on theSecond International Conference on MultiagentSystems (ICMAS’96) Mobile Assistant Project thatwas conducted at an actual international confer-ence for multiagent systems using 100 personaldigital assistants (PDAs) and cellular telephones.We supported three types of service: (1) communi-cation services such as e-mail and net news; (2)information services such as conference, personal,and tourist information; and (3) community sup-port services such as forum and meeting arrange-ments. After the conference, we analyzed a largeamount of log data and obtained the followingresults: It appears that people continuously usedPDAs in their hotel rooms after dinner; e-mail ser-vices were used independently of the conferencestructure, but the load on information servicesreflected the schedule of the conference. Postques-tionnaire data showed that our trial was consid-ered interesting, although people were not fullysatisfied with the PDAs and services provided. Par-ticipants showed a deep interest in mobile com-puting for community support.

The project reported in this article isintended to support, using mobile com-puting, increasingly diverse and amor-

phous groups of people. For this purpose,mobile computing services should change fromapplications based on point-to-point communi-cation to those based on communitywide com-munication. Community mobile computing isfeasible because the number of users of wirelesscommunication services is rapidly increasingbecause of the falling prices of mobile terminalsand cellular telephones. We show that mobilecomputing technology supports internationalconferences effectively and combines well withconventional desktop computing technology.

Internet access services are now popular atinternational conferences. Several tens of desk-

top terminals are usually provided at a confer-ence site. However, the services are constrainedin both space and time; the services are onlyaccessible from terminal rooms during the day-time. However, the salient feature of mobilecomputing is that users can use their personaldigital assistants (PDAs) anytime and anywhere.The challenge of this article is to apply mobilecomputing to an actual international confer-ence and investigate its role quantitatively.

Previous work in computer-mediated com-munication was classified into two main cate-gories (Mynatt et al. 1997): First is multiuserdungeons (MUDs) and casual meeting toolsthat support a virtual space where everybodycan meet and talk. The other is MEDIASPACE (Bly,Harrison, and Irwin 1993), which is a hybrid ofvirtual and real spaces. Mobile computing,however, aims to amplify interactions in a realspace and should therefore be classified in athird category.

Although internet access services are com-mon at international conferences, no empiricalstudy on them has been reported. Similarly, inthe mobile computing literature, several papershave been published concerning technologiesand applications, but no analysis on socialinteractions has been published (Weiser 1993).Human-computer interaction (HCI) researchershave expended substantial effort on multime-dia communications and conferences for rela-tively small teams (Olson, Olson, and Meader1995). Several reports address large groups inbusiness applications (Sproull and Kiesler1988), education (O’Day, Bobrow, and Shirley1996), and home computing (Kraut et al. 1996).However, there is no report on mobile comput-ing with a large number of users.

In this article, we first report the experimentcalled the Second International Conference onMultiagent Systems (ICMAS’96) Mobile Assis-

Articles

SUMMER 1998 31

Mobile Digital Assistants for Community Support Yoshiyasu Nishibe, Hiroaki Waki, Ichiro Morihara, Fumio Hattori,

Toru Ishida, Toshikazu Nishimura, Hirofumi Yamaki, Takaaki Komura,Nobuyasu Itoh, Tadahiro Gotoh, Toyoaki Nishida, Hideaki Takeda,

Atsushi Sawada, Harumi Maeda, Masao Kajihara, and Hidekazu Adachi

Copyright © 1998, American Association for Artificial Intelligence. All rights reserved. 0738-4602-1998 / $2.00

AI Magazine Volume 19 Number 2 (1998) (© AAAI)

The goal of the

ICMAS’96Mobile

Assistant Project wasnot only to

support communi-

cation services such as

e-mail butalso to

provide,throughPDAs, a

variety ofinformationrequired at

internationalconferences.

running on MAGICLINK.2 The server system wasprogrammed with TELESCRIPT (White 1994),2 alanguage intended for mobile computingapplications. The client system was built onMAGICCAP OS running on MAGICLINK. Each clientcould connect with the server system used bywireless public telephone lines. The data-trans-mission rate of wireless lines on the physicallayer was 9600 bits a second (bits/s).

The project became larger than we firstexpected. Thirty telephone lines were newlyintroduced and connected with the servermachine placed at the conference site. Wethen found a serious communication problembetween the PDAs and the server. Becausemobile communication traffic in this experi-ment differed from typical telephone services,we found that the existing facilities were notenough to cover the demand. After negotia-tion with the local cellular telephone compa-ny, the situation was substantially improvedby allowing several tens of calls from the con-ference site to be conducted simultaneously.

The project was announced to conferenceattendees beforehand so that they could applyfor a PDA. We asked all participants to fill outa prequestionnaire together with the applica-tion form. At the conference registration desk,after the process of authentication and con-tract, the PDAs and the telephones were hand-ed to the project participants.

Figure 3a shows a participant receiving aclient system at the registration desk. The con-tract included permission to use log data foracademic purposes. During the conference, ahelp desk was set up at the conference site toassist participants with various technical prob-lems. Figure 3b depicts a participant using theservice while he stands in the lobby of the con-ference hall. Figure 3c also depicts a partici-pant using the service in the event hall wherethe conference was held. Figure 3d depicts thehelp desk. Figure 3e is a photograph taken dur-ing the excursion to Nara Historical Park,which was done as a part of the conferenceschedule. Help-desk staff also traveled to NaraPark during the excursion. After the confer-ence, all PDAs and cellular telephones werereturned without trouble.

To design the mobile digital assistant as anagent for community support, we first classifiedan individual’s activities into the followingthree categories: (1) communicating with otherpeople to share information, (2) interactingwith the real world to determine his/heractions, and (3) understanding communityactivities to identify his/her role in the commu-nity. We then designed various functions tosupport these three types of activity and imple-

tant Project. This conference was held inKyoto, Japan, from 9 to 13 December 1996.The project provided (1) e-mail and internetaccess services; (2) conference, personal, andtourist information services; and (3) forum andmeeting arrangement services. In this experi-ment, about 100 personal digital assistants(PDAs) with wireless telephones were loanedto conference attendees without any charge totry out the system. To the best of our knowl-edge, this was the world’s first application ofmobile computing to community support.

By reducing the time and space constraintsof desktop computing, mobile computingguarantees freedom of use of network services.As a result, the demand for various informa-tion services was clearly revealed. We showthat mobile-computing technology supportsinternational conferences effectively and com-bines well with conventional desktop comput-ing technology.

ICMAS’96 Mobile Assistant Project

The goal of the ICMAS’96 Mobile AssistantProject1 was not only to support communica-tion services such as e-mail but also to provide,through PDAs, a variety of informationrequired at international conferences (figure1). PDAs were provided to about 100 people,roughly one-third the conference attendees.People could use them in the conference site,lobbies, and hotel rooms as well as outdoors.

Figure 2 describes an overview of the systemconfiguration. The server machine was con-nected to the internet to provide e-mailexchange and information retrieval. The serversystem was programmed with TELESCRIPT, whichis a language intended for agent-orientedmobile computing applications (White 1994).The client system was built on MAGICCAP OS

Articles

32 AI MAGAZINE

Figure 1. PDA with Telephone.

mented (1) communication services, (2) infor-mation services, and (3) community services, asdescribed in table 1. These services consisted ofcommercial services such as e-mail and variousoriginal services developed by about 20 engi-neers and students from NTT Information andCommunication Systems Laboratories, KyotoUniversity, and the Nara Institute of Scienceand Technology. The original services given intable 1 were provided during the conference:

First, ACTION NAVIGATOR is a system that sup-plies tourist information and additional infor-mation to help the user make decisions.

Second, INFOCOMMON provides a weak infor-mation structure for human information shar-ing. The weak information structure connectsvarious information media without definingthe semantics rigorously.

Third, COMMUNITY VIEWER visualizes interac-tions among participants and provides a betterview of community activities. Figure 4 showsthe screen image of the opening index forselecting these services. Because most serviceswere developed just for this experiment, wewere able to embed enough codes to obtain logdata such as the log for access to the server andchanges in PDA output. We also asked all usersto fill out a prequestionnaire and a postques-tionnaire to find out about their usual comput-er environments and impressions of the pro-ject and the services provided.

ACTION NAVIGATOR: Supporting Action MakingWe are always making decisions, such as select-ing the restaurant when we eat out (Engel,Blackwell, and Miniard 1990). Typical touristinformation services provide objective infor-mation such as name, telephone number,address, and business information when usersrequest information.

However, depending on search conditions,they often return a large number of searchresults. In this case, the user has to choose theone he/she wants from this large amount ofinformation. It is difficult to choose on thebasis of objective information only. In general,when we make decisions, we often refer notonly to objective information but also to sub-jective information, for example, recommen-dations of others or reviews in newspapers ormagazines.

ACTION NAVIGATOR (Ohtsubo et al. 1998) sup-ports decision making with subjective infor-mation from others coupled with a touristinformation service. Figure 5 shows an exam-ple screen of ACTION NAVIGATOR. Informationspots are displayed in the map of Nara city.Users could get detailed information on shops,restaurants, and so on, by clicking on the cor-responding spot.

The subjective information in ACTION NAVIGA-TOR was based on answers to the following two

Articles

SUMMER 1998 33

The Internet

Server SystemTelescript Engine

PublicTelephone Network

30 Lines

Terminal Server(ANNEX-III)

Client System(100 Participants)

HP9000 model 800/I60

MagicLink 2100JPlus Digital Mova MagicLink 2100J

Plus Digital Mova

MagicLink 2100JPlus Digital Mova

MagicLink 2100JPlus Digital Mova

MagicLink 2100JPlus Digital Mova

Figure 2. System Configuration.

levels and changes the size of each informa-tion spot according to these levels. The activelevel took four values in this experiment. Theinformation spot that has the highest activelevel (displayed as the biggest icon) is calledthe hot-spot.

We examined the accesses to the informa-tion spots. It appears that 50 percent of allaccesses (furthermore, 80 percent of the firstaccesses) were to the top 7 percent of the hot-spots. The results show that the simple hot-spot mechanism effectively realizes social fil-tering of tourist information.

However, users did some decision making

questions: (1) How often do users look at thedetailed information for each informationspot? (2) Do users post news about this infor-mation spot? Each PDA counts the number oftimes that the user refers to the detailed infor-mation and sends this count to the server. Theserver gathers the number of references andcalculates an evaluation value (we call it theactive level) for each information spot.

When starting up ACTION NAVIGATOR, users areasked whether or not the system shouldupdate the active levels. When the user choos-es to update these levels, the PDA communi-cates with the server to get the current active

The subjective

informationin ACTION

NAVIGATOR

was based onanswers to thefollowing two

questions:

(1) How oftendo users look at

the detailedinformation

for each information

spot?

(2) Do userspost newsabout this

informationspot?

Articles

34 AI MAGAZINE

Figure 3. System in Use.A. Registration desk. B. Lobby. C. Event hall. D. Help desk. E. Nara Park.

A B

E

DC

Communication servicesSend message Messages are sent from PDAs to other participants or

internet users through telephone lines.Receive message Messages are received by PDAs from other participants

and internet users through telephone lines.

Information servicesConference information Session schedules and maps are stored in each PDA.

Abstracts of presentations are provided through telephone lines.

Personal information COMMUNITY VIEWER provides participants’ personal information stored in each PDA and visualizes theinteractions among the participants through telephone-line transfer.

Tourist information ACTION NAVIGATOR provides information (stored in each PDA) such as shops and sightseeing spots.

Community servicesForum and meeting INFOCOMMON helps users to exchange knowledge and

ideas through shared card information through telephone lines. SOCIAL MATCHMATER provides a way to find other people who have similar interests.

Statistics feedback The current state of service use is shown in real time through telephone lines.

PDA = personal digital assistant.

Articles

SUMMER 1998 35

Table 1. Services Provided.

Figure 4. Opening Index.

others takes an important role in making deci-sions. The result supports our policy of design-ing ACTION NAVIGATOR.

INFOCOMMON: Sharing Information among ParticipantsThe effective use of tacit knowledge or nonlin-guistic means of communication would be an

during the conference independent of usingACTION NAVIGATOR. We asked users the followingquestion in the postquestionnaire: “If youwent to some restaurants or sightseeing, whydid you select them?” Figure 6 shows theanswers to the question. According to thischart, users’ decisions depended on the recom-mendations of others: Individual preference of

Articles

36 AI MAGAZINE

Figure 5. Screen Image of ACTION NAVIGATOR.

Information Taken by Action Navigator

Recommendationsof Others

Invitations of Others

Other0

10

20

30

Nu

mbe

r of A

nsw

ers

6(11%)

26(46%)

18(32%)

6(11%)

Figure 6. The Factor of Making Decisions.

interesting issue in addition to unambiguousrepresentation of information. It is not untilrecently that these issues have been addressedfrom computational points of view.

The basic assumption behind INFOCOMMON

(Maeda et al. 1997) is that information sharingfacilitates human interaction and group for-mation (Gaines and Shaw 1994). Talking abouta common hobby might result in an academicdiscussion. To facilitate information sharingamong participants, we used several designprinciples: First, we tried not to enforce anyone particular concept. Instead, we allowed alot of freedom in the use of terms and thestructure of shared information and freedomto incorporate useful information from variousviewpoints. Second, we made the informationspace a single seamless space, which releasesthe user from working with a rigid menu.Third, we enabled the user to build a personalinformation space where he/she can organizerelatively small amounts of information, asdesired (Nishida et al. 1995). This approach,we believe, makes shared information visibleand encourages human interaction and com-munity formation.

Technically, we propose a weak informationstructure for encoding shared information.The weak information structure connects awide variety of information media, such as

natural language text, hypertext, and images,without defining the semantics rigorously. Byleaving ambiguity and vagueness in the repre-sentation, we can incorporate a wider varietyof useful information into the informationbase, ranging from formal conference informa-tion to articles that might be useful in findinginteresting restaurants or points of interest.

The information base of INFOCOMMON is acollection of information cards organized onthis weak information structure. Representa-tive keywords are automatically extracted fromeach information card. The collection of infor-mation cards is organized by shared keywordsand predefined relations. Given a set of key-words, INFOCOMMON responds with the set ofinformation cards connected to the keywordsand provides a visual interface for retrievingand sending information cards. Figure 7 showsan example screen of INFOCOMMON. The rela-tion between two information cards is dis-played by a link. The result of retrieval is storedin the user’s local information base where theuser can rearrange the collection of informa-tion cards and add and remove nodes andlinks, as desired.

We analyzed how INFOCOMMON was used bylog files and questionnaires. Figure 8a showsthe answers to the question, “For what did youused INFOCOMMON?” Fifty-nine percent used the

Fifty-nine percent usedthe system forinformationretrieval and19 percent forinformationsending; sending comprisesnews, help desk, and discussion.

Articles

SUMMER 1998 37

Figure 7. Screen Image of INFOCOMMON.

ventional news reader. Major reasons for thechoice “didn’t use” were “slow informationretrieval” (8 persons) and “I couldn’t connectto the server” (4 persons). These problemsinvolve server response, so they can be correct-ed easily.

Figure 8b displays answers to the question,“For what topic did you use INFOCOMMON?”About search results, 81 percent felt that the

system for information retrieval and 19 per-cent for information sending; sending com-prises news, help desk, and discussion. The rea-sons INFOCOMMON was used for news “becausekeyword search was easy and useful” (14 per-sons), “because I found interesting topics inINFOCOMMON” (5 persons), and “because I had aquestion” (4 persons). We determined thatINFOCOMMON added a new facility to the con-

Articles

38 AI MAGAZINE

InformationRetrieval

HelpDesk

Discussion News Others NotUsed

0

10

20

30

40

50

60

Nu

mbe

r of

An

swer

s 49(59%)

6(7%)9(11%)

1(0%) 0(0%)

18(22%)

Research Presen-tation

Sight-seeing

Restau-rant

People Others0

10

20

30

40

50

60

Nu

mbe

r of

An

swer

s

24(33%)

13(18%)11(15%)

8(11%)12(16%)

5(7%)

Figure 8. Use of INFOCOMMON.A. Purpose. B. Topic.

A

B

search results were fine, and 51 percentanswered that they were satisfied with INFO-COMMON. About the usefulness of INFOCOMMON

for information retrieval and discussion, 55percent answered that INFOCOMMON was usefulfor getting the information they needed. How-ever, only 26 percent answered that it was use-ful for discussion. We feel that the five-dayperiod was too short to form the kind of com-munity in which people create many activediscussions. We need to conduct a longer-termexperiment to evaluate the usefulness of thesystem for discussion.

COMMUNITY VIEWER: Visualizing Community ActivitiesCOMMUNITY VIEWER (Nishimura et al. 1998) pro-vides the following three types of information:(1) personal profiles of conference participantsand their interests; (2) static relations amongthe participants, as derived from the personalprofiles; and (3) dynamic activities of the par-ticipants, including the ongoing process ofcommunity formation. The goal of COMMUNITY

VIEWER is to support the formation of commu-nities wherein people are willing to share theirpersonal interests. At the first stage of commu-nity formation, personal profiles can help peo-ple in deciding whom to share personal inter-ests with. Sharing knowledge (to know whatother people know) and activities (to do whatother people do) is also useful. Thus, the ques-tion is how to increase mutual interests,knowledge, and activities without infringingon people’s privacy. This dilemma became aserious problem in designing COMMUNITY VIEW-ER because private communication takes amore important role in community supportsystems than in traditional groupware.

These types of information are provided in aunified framework called the party room, whichis a virtual place for visualizing the activities ofthe community. To enhance the awareness ofthe community among the participants andprotect their privacy, we introduced the newconcept of reflector icon. We can observe thedynamic activities in the community by glanc-ing over the room. Our party room has twodisplay modes: (1) the overview mode thattakes in the whole room and (2) the detailedmode that observes the activities of each indi-vidual. Figure 9a shows a typical screen imagein the detailed mode.

Each icon can represent abstract informa-tion about personal activities without infring-ing on personal privacy. In the party roommetaphor, the reflector icon is realized as a facemark (figure 9). One icon represents one par-ticipant of the community. In COMMUNITY VIEW-

ER, people can easily access personal profiles byselecting reflector icons. A detailed profile isshown in figure 9b.

Several functions are provided to distinguishamong individuals in the party room. Supposeyou want to find those who work in researchfields similar to yours. You can do this by spec-ifying the appropriate keywords as search con-ditions. When the system finds the other par-ticipants who match the conditions, the colorof the corresponding reflector icons change,and the icons move toward your icon. As aresult, you can easily access the personal infor-mation as needed and determine to whom tosend a meeting request.

The ongoing interactions in the communityare visualized in the party room. The behaviorof each icon reflects the current activity of thecorresponding person in the following man-ner: In the default state, when a user is notinteracting, his/her icon walks toward a ran-domly selected table to sit down in the partyroom. After staying at the table for a time, theicon leaves for a different table. When oneinteracts (sends a message, reads a personalprofile, and so on) with others in the virtualparty room, the corresponding icon behaves asif approaching and talking to others. As thedegree of interaction increases, icons start tocongregate, mirroring what is seen at parties.The length of time over which the icons asso-ciate is proportional to the degree of interac-tion between the people. Thus, users get theimpression that people in the community areactively interacting, although the contents ofthe interactions are not displayed.

To realize this function, the system has tojudge who is interacting with whom and towhat degree. In the experiments, we used thenumber of accesses to the personal profile asthe degree of interaction. In general, e-mail,telephone calls, meeting arrangement requests,and any other interaction can be considered toindicate the degree of interaction. Each PDAcalculates how much its owner is interested inothers and sends the information to the com-munity server. The server accumulates all theinformation and sends a summary back to eachPDA. To visualize personal characteristics in acommunity, the reflector icon changes its sizeaccording to run-time statistics. People canknow who is attracting attention and who iscurrently active by observing icon size.

Interaction AnalysisThe challenge in this section is to investigatequantitatively the role of mobile computing inan actual international conference. To analyze

The goal ofCOMMUNITY

VIEWER is tosupport theformation ofcommunitieswherein people arewilling toshare theirpersonalinterests.

Articles

SUMMER 1998 39

Articles

40 AI MAGAZINE

A

B

Figure 9. Screen Image of COMMUNITY VIEWER.

A. Party room. B. Personal profile.

a large amount of noisy data, we were carefulin selecting key data items. Our strategy foranalysis of log data was to select data itemsthat were independent of the provided soft-ware functions and implementations. In addi-tion, we excluded the log data of project mem-bers and system operators to get accurate useresults from true participants.

Our first goal was to determine how mobilecomputing is used in conferences and how itsuse differs from that of desktop computing.Figure 10 represents the overall activities dur-ing the conference. The x-axis indicates thehour, and the y-axis indicates the number ofevents within this hour. The first two days(9–10 December) were for workshops and tuto-rials, and the next three days (11–13 Decem-ber) were for technical programs. The recep-tion was held on the evening of 11 December,and the excursion to Nara Park was set for theafternoon of 13 December.

The following features can be observed infigure 10:

Continuous use: Except for the receptionand excursion, the system was used continu-

ously, even during technical presentations.People sometimes retrieved related informa-tion using PDAs while they listened to a pre-sentation.

Midnight use: People used PDAs even afterthe conference, especially in their hotels afterdinner. The highest peak in use, at midnight on10 December, shows that PDAs are activelyused just before the technical program starts.

Figure 11 shows the data for traffic on thethird day, which showed the most typical traf-fic characteristic of the project. For purposes ofcomparison, it also shows the number of tele-phone calls made in one day on PASEO (Niwanoet al. 1997), a multimedia communicationmarket trial service provided by NTT FutureAgent Network Inc., which also used MAGIC-LINK, the same PDA as in our experiment.

In PASEO, traffic peaked at 9:00 AM, noon, and8:00 PM. These traffic peaks were because mostusers on PASEO had difficulty accessing the serv-er during business hours. In the mobile assis-tant project, however, traffic dropped on theevening of the third day and the afternoon ofthe fourth day. These periods match the times

Articles

SUMMER 1998 41

ReceptionSessionSession Session Session Session

Excursion

0

10

20

30

40

50

60

70

80

90

Date/Time

0:0012:00 0:0012:00 0:0012:00 0:0012:00 12:00

12/09 12/10 12/11 12/12 12/13

Nu

mbe

r of

Eve

nts

Figure 10. Use of Mobile Computing.

at the beginning of the conference and gradu-ally decreased toward the end. However, astructural trend exists as follows:

Demand for e-mail services was fairlysteady during the conference. The number ofrelated events (sending and receiving mes-sages) was not dependent on the progress ofthe conference but, rather, on the number ofPDAs in use.

However, the demand for information ser-vices changed dramatically. Conference infor-mation services were used most on 11 Decem-ber, the start of the technical programs. Thedemand for personal information services washigh at the beginning of the conference andslowly decreased thereafter.

Although the number of accesses was notlarge, the demand for statistics feedbackshowed a trend similar to that of personalinformation services: active use at the begin-ning of the conference with a slow drop-offthereafter.

Figure 12b shows the daily trend of partici-pants’ activity. The x-axis indicates three-hourblocks, and the y-axis indicates the sum of

for the conference reception and a postconfer-ence excursion, during which most users didnot use their PDAs. Furthermore, the trafficpeaked at midnight on each day. Except duringthe reception and excursion and in the veryearly morning, the system was used continu-ously. In this experiment, users could use theirPDAs whether they had connected to the serv-er, even during others’ technical presentations.Traffic patterns on public services correlatewith such things as hours of business, and thetraffic pattern in this experiment correlateswith the conference schedule. Thus, welearned that the use of a mobile computingsystem strongly reflects the users’ situation.

Use of Services We then analyzed the demands in internation-al conferences by investigating how the ser-vices were used during the conference. Figure12a describes the week’s trend of participants’activity. The activity was low on 9 and 13December because the PDAs were picked upand returned on these two days. At first glance,except for the first two days, activity was high

Articles

42 AI MAGAZINE

ReceptionConference

m

m

m m m

m m

m

m

m

m

m

m

m

m

m

m

m

m

m

mm

m

m

mo

o

o

o

oo o

o o

o

o

o

oo o

o

o o

oo

o

o

o

oo

0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:000

10

20

30

40

50

60

70

80

Time

m This Experiment

o Paseo

Nu

mbe

r of

Acc

esse

s

Figure 11. Number of Telephone Calls (comparison with PASEO).

Articles

SUMMER 1998 43

Dec. 9 Dec. 10 Dec. 11 Dec. 12 Dec. 130

100

200

300

400

500

600

700

800

900

Figure 12. Use of Services.

A. Trend over the Entire Week. B. Daily Trend.

0

100

200

300

400

500

600

0:00

- 3

:00

3:00

- 6

:00

6:00

- 9

:00

9:00

- 1

2:00

12:0

0 -

15:0

0

15:0

0 -

18:0

0

18:0

0 -

21:0

0

21:0

0 -

24:0

0

Statistic FeedbackForum and MeetingTourist InformationPersonal InformationConference InformationReceive MessageSend MessageA

B

Connection RefusalWe also tried to find the effects of the transmis-sion rate and communication errors on mobilecomputing services. For e-mail, figure 13shows the data-transmission rate.

In figure 13, points marked with an asteriskmean values for which the transmission hadan incomplete status because of errors or userinterruptions. In figure 13, the most impor-tant characteristic is the distribution of thedata-transmission rate. Typically, the data-transmission rate on services using a wiredlocal area network follows a normal distribu-tion. The data-transmission rate in this exper-iment, however, has an offset distribution.Almost all measured points are concentratedaround the line that defines an average data-transmission rate of 400 bits/s. Some pointsappear on the higher-rate side, but a fewpoints appear on the lower-rate side. An aver-age data-transmission rate of 400 bits/s is lowcompared with the 9600 bits/s of the physicallayer on our mobile computing system. Fromthese results (the low transmission rate andconcentrated distribution), we estimate thatin most cases of transmission in this experi-ment, the phenomena that caused a low-rateof transmission were transmission delay, biterror on the transmission layer, the retrans-mission of packets, and so on.

events that occurred within the correspondingtime interval during the five days. The majorresults obtained from this figure are as follows:

Activity increased from morning to night.Lunch time did not affect the users’ behavior,but dinner time (reception and excursion)substantially decreased their activity. Aremarkable peak appears from dinner to mid-night. E-mail services were used steadily allday. Messages from the internet were receivedall night.

Conference information was retrieved fre-quently in the morning and after dinner, butpersonal information was often retrieved inthe afternoon and particularly after dinner.

In summary, the demand for e-mail ser-vices was steady, but that for information ser-vices was heavily dependent on the confer-ence structure. In previous conferencesupport, desktop computing primarily pro-vided e-mail services so that users could con-tinue their business at the conference site.Mobile computing, however, can supportmore conference-oriented information ser-vices. The results of log data analysis createthe picture of people using their PDAs afterdinner in their hotel rooms to get additionalinformation on other people whom they metin the afternoon and make an action sched-ule for the next day.

Articles

44 AI MAGAZINE

✱

✱✱

✱✱✱✱✱

✱

✱

✱✱

✱✱

✱

✱✱

✱✱

❍✱

❍

❍

❍

❍

❍

❍❍

❍

❍

❍

❍❍❍❍

❍❍❍

❍

❍

❍

❍

❍

❍

❍

❍

❍

❍

❍❍

❍❍

❍

❍

❍

❍❍

❍

❍

❍

❍❍

❍

❍

❍

❍

❍

❍❍

❍❍

❍

❍

❍

❍

❍❍❍❍

❍

❍

❍

❍

❍

❍❍❍

❍

❍❍

❍

❍

❍

❍

❍❍

❍

❍

❍

❍

❍

❍

❍

❍

❍❍❍

❍

❍❍

❍

❍

❍

❍❍

❍

❍

❍❍

❍❍

❍❍

❍

❍❍

❍❍❍

❍

❍❍❍

❍

❍

❍

❍

❍❍

❍

❍

❍

❍❍

❍

❍❍

❍❍

❍

❍

❍❍

❍❍

❍❍

❍

❍

❍❍

❍

❍

❍

❍❍❍

❍

❍❍

❍

❍

❍

❍

❍

❍

❍❍❍

❍❍❍❍

❍

❍

❍

❍

❍❍❍

❍

❍❍❍

❍

❍

❍❍❍❍❍

❍❍❍

❍

❍

❍

❍

❍❍

❍❍

❍

❍❍

❍

❍

❍❍

❍❍❍

❍

❍

❍❍❍❍❍❍❍❍❍

❍

❍❍❍❍

❍

❍❍

❍

❍❍❍❍❍

❍

❍❍❍❍❍❍❍❍❍

❍❍❍❍❍❍❍

❍❍❍

❍❍

❍

❍❍❍❍❍❍❍❍❍❍❍❍

❍

❍❍

❍❍❍❍❍

❍❍

❍❍❍❍❍❍

❍

❍

❍

❍❍❍❍❍❍

❍❍❍❍❍

❍

❍❍

❍

❍❍❍❍❍❍

❍❍❍❍

❍

❍❍❍❍❍❍❍❍❍❍❍❍

❍

❍

❍❍❍❍❍❍❍

❍❍❍❍❍❍❍❍❍❍❍

❍

❍

❍❍❍

❍❍❍❍

❍

❍❍

❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍

❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍

❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍

❍

❍❍❍

❍

❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍

❍

❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍

❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍

✱✱

✱

✱✱✱

✱

✱

✱

✱✱✱✱✱

✱

✱

✱

✱✱✱✱

✱

✱

✱✱✱✱

✱

✱

✱✱✱✱✱

0

50

100

150

200

250

300

350

400

450

500

0 10 20 30 40 50 60 70 80 90

Tim

e [s

]

Data Size [kbytes]

✱ Incomplete Transmission❍ Complete Transmission

✱

Figure 13. Data Transmission.

Articles

SUMMER 1998 45

Comments from Participants

■ It is a great initiative and I think it is the basis for the future mobile agent. I

sometimes found it cumbersome; in general, if the communicational aspects

improve, it will be of great value.

■ It’ll be more interesting if the participants use it more often. Less info, less news,

less posting is not interesting.

■ In general, the project is interesting and seems to have many possibilities. The

concrete realization has many difficulties and problems. Besides those already

mentioned, there have to be people doing first steps, i.e., posting news, inviting

other people to discussion, meetings, etc. Without proving this initial start by

yourselves, I do not think that much action will happen (I am also to blame, but

it seems that nobody did anything, so I am not the only one.)

■ I think the project is very interesting and potentially very important. The soft-

ware should be enhanced, as well as the quality of the information, and the way

it could be modified. The design of the user interface should be changed; it is

not very clear how to use it.

■ Unfortunately the interaction speed/ease was far too slow for me to use. It was

much easier just to meet people and I could never get e-mail to work. I did not

have time to visit the help desk to see what was wrong. Good idea but the

speed wasn’t useful for me.

■ The project was nicely arranged and I appreciated being a part of it. I was very

disappointed with the PDA for several reasons. First, I had to have a special

address that people at home did not know about. Second, I really, really, missed

standard tools like TELNET and FTP. Finally, If you were taking notes on talks, the

only thing you could do with them was send them home by e-mail. It would

have been nice to be able to create files that could be transferred, through FTP,

to your home server.

structure, but the peak of information servicesdepended on the progress of the conference.

Important for the design of information ser-vices on mobile computing, we also found thatusers had a tendency to break the connectionwhen they were forced to wait a long time fordata transmission. Most users who had someerror or interruption in transmission recon-nected to the server system after a relativelylong time.

We know that a single trial is not enough toobtain general conclusions. However, it is alsotrue that large-scale experiments are not easyto repeat. To the best of our knowledge, noanalytic reports have been published on con-ference support services, even for desktopcomputing. We believe it is valuable to sharethe results of our log data analysis with otherresearchers, who might perform similar exper-iments in the future.

Postquestionnaire data showed that our trialwas considered interesting, although peoplewere not fully satisfied with the PDAs and theservices provided. (See the sidebar for moredetailed comments from participants.) The par-ticipants showed a deep interest in mobile com-puting for convention support. This project wasplanned as a step toward a new research initia-tive on communityware (Ishida 1998).

Systems that are to be used in a communityare more difficult to design than point-to-pointcommunication services. They cannot be eval-uated by the simple sum of the impression ofthe users. The effectiveness of such systems

In questionnaires, users claimed that thedata-transmission rate of communication withthe server system was slow (transmission rate =400 bits/s and setup time = 29 s; negotiationbetween modems = 23 s, and the time to estab-lish PPP [point-to-point protocol] = 6 s).

Figure 13 shows that user interruptionsoccurred during the early stages of transmis-sion. In this experiment, the total size of datato be transmitted and the amount transmittedto this point were shown on the client system.Thus, users were able to estimate the total timeof transmission. If the user knew that transmit-ting data from the server would take a longtime, he/she often interrupted the transmis-sion.

ConclusionsThrough the ICMAS’96 Mobile Assistant Pro-ject, we provided digital mobile assistants asagents for community support. The systemwas actually tested at an international confer-ence, and the data collected are being ana-lyzed.

The analysis of user behavior clearly illus-trates the role of mobile computing in support-ing international conferences. PDAs were con-tinuously used not only at the conference sitebut also elsewhere after dinner. People usedPDAs more often than desktop computing butfor short periods only. E-mail services were usedsteadily and independently of the conference

Articles

46 AI MAGAZINE

very much

almostnot much

not at all

very muchalmost

not much

Figure 14. Questionnaire.A. Satisfaction with services. B. Agreement to trials.

must be verified through experiments involvinglarge numbers of people. Through real experi-ments, we expected to confirm what the role ofmobile computing in social interactions amonghuman communities is. Figure 14a indicates thedegree of satisfaction with the provided ser-vices, and Figure 14b indicates the degree ofagreement with this trial. It is notable that theproject was widely accepted, although we couldnot completely satisfy all participants. Peopleclaimed that the terminal was heavy and incon-venient to carry, the speed of communicationwas slow, and so on. However, we can see thatthe users showed a deep interest in mobile com-puting for community support.

AcknowledgmentsThe ICMAS’96 Mobile Assistant Project wasconducted collaboratively by NTT Informationand Communication Systems Laboratories,Kyoto University, Nara Institute of Science andTechnology, and Kobe University and in coop-eration with NTT Kansai Mobile Communica-tions Network Inc., General Magic Inc., NTTMultimedia Service Promotion Headquarters,and Keihanna Interaction Plaza Inc. We wantto thank Katsumi Tanaka, Shiro Takada, RieOhtsubo, and Katsumi Takahashi for their helpin this project. We would like to thank the par-ticipants of this project for their cooperationand encouragement.

Notes1. A more detailed project report can be found inNishibe et al. (1998).

2. Now, agent-oriented mobile computing languagesare mainly based on JAVA, for example, AGLETS andODYSSEY.

ReferencesBly, S.; Harrison, S.; and Irwin, S. 1993. Media Spaces:Bringing People Together in a Video, Audio, andComputing Environment. Communications of theACM 36(1): 28–47.

Engel, J. F.; Blackwell, R. D.; and Miniard, P. W. 1990.Consumer Behavior. 6th ed. Fort Worth, Tex.: Dryden.

Gaines, B. R., and Shaw, M. L. G. 1994. Using Knowl-edge Acquisition and Representation Tools to Sup-port Scientific Communities. In Proceedings of theTwelfth National Conference on Artificial Intelli-gence, 707–714. Menlo Park, Calif.: American Asso-ciation for Artificial Intelligence.

Ishida, T., ed. 1998. Communityware: Concept andPractice. New York: Wiley.

Kraut, E. R.; Scherlis, W.; Mukhopadhyay, T.; Man-ning, J.; and Kiesler, S. 1996. HOMENET: A Field Trial ofResidential Internet Services. In Proceedings of theInternational Conferences on Human Factors inComputing Systems, 284–291. Vancouver, B.C.,Canada: International Conferences on Human Fac-tors in Computing Systems.

Maeda, H.; Kajihara, M.; Adachi, H.; Sawada, A.;Takeda, H.; and Nishida, T. 1997. Weak InformationStructures for Community Information Sharing.International Journal of Knowledge-Based IntelligentEngineering Systems 1(4): 225–234.

Mynatt, E. D.; Adler, A.; Ito, M.; and O’Day, V. L.1997. Design for Network Communities. In Proceed-ings of the International Conferences on HumanFactors in Computing Systems, 210–221. Atlanta,Ga.: International Conferences on Human Factors inComputing Systems.

Nishibe, Y.; Morihara, I.; Hattori, F.; Nishimura, T.;Ishida. T.; Maeda. H.; and Nishida, T. 1998. MobileDigital Assistants for International Conferences. InCommunityware: Concept and Practice, ed. T. Ishida,245–284. New York: Wiley.

Nishida, T.; Takeda, H.; Iino, K.; and Nishiki, M.1995. A Knowledge Media Approach to OntologyDevelopment. In Toward Very Large Knowledge Bases:Knowledge Building, Knowledge Sharing, ed. N. J. I.Mars, 84–94. Amsterdam, The Netherlands: IOS.

Nishimura, T.; Yamaki, H.; Komura, T.; Itoh, N.;Gotoh, T.; and Ishida, T. 1998. COMMUNITY VIEWER:Visualizing Community Formation on Personal Dig-ital Assistants. Paper presented at the 1998 ACMSymposium on Applied Computing (ACM SAC’98),Atlanta, Georgia. Forthcoming.

Niwano, E.; Okamoto, K.; Sakanoue, K.; Otaka, H.;Katsumata, M.; and Maeda, K. 1997. Evaluation ofLarge-Scale Mobile Multimedia Communication Sys-tem Built on Telescript Agent Platform—The PASEO

Service. In Proceedings of IEEE Multimedia 97.Washington, D.C.: IEEE Computer Society.

O’Day, V. L.; Bobrow, D. G.; and Shirley, M. 1996.The Social-Technical Design Circle. In Proceedings ofthe International Conference on Computer-Support-ed Cooperative Work, 160–169. Boston, Mass.: Inter-national Conference on Computer-Supported Coop-erative Work.

Ohtsubo, R.; Takahashi, K.; Nishibe, Y.; and Mori-hara, I. 1998. ACTION NAVIGATOR: An Information Ser-vice Based on Agent Communication for SupportingDecision Making. In Proceedings of the Third Inter-national Conference on the Practical Application ofIntelligent Agents and Multiagent Technology(PAMM-98). London: International Conference onthe Practical Application of Intelligent Agents andMultiagent Technology.

Olson, J. S.; Olson, G. M.; and Meader, D. K. 1995.What Mix of Video and Audio Is Useful for SmallGroups Doing Remote Real-Time Design Work? InProceedings of the International Conferences on HumanFactors in Computing Systems, 362–368. San Francisco,Calif.: Morgan Kaufmann.

Sproull, L., and Kiesler, S. 1988. Reducing SocialContext Clues: Electronic Mail in OrganizationalCommunication. In Readings in Computer-SupportedCooperative Work, 684–712. San Francisco, Calif.:Morgan Kaufmann.

Weiser, M. 1993. Some Computer Science Issues inUbiquitous Computing. Communications of the ACM36(7): 74–83.

White, J. 1994. Telescript Technology: The Founda-

Important for the design ofinformationservices onmobile computing,we also found thatusers had atendency tobreak the connectionwhen theywere forcedto wait a

long time for data transmission.

Articles

SUMMER 1998 47

Toshikazu Nishimura is currentlyan assistant professor in theDepartment of Social Informaticsat Kyoto University. He receivedhis B.E. and M.E. in 1990 and1992, respectively, from KyotoUniversity. His research interestsinclude computer-mediated com-munication and mobile comput-

ing systems. His e-mail address is nisimura@kuis.kyoto-u.ac.jp.

Hirofumi Yamaki is currently agraduate student in a Ph.D. pro-gram in the Department of Infor-mation Science at Kyoto Universi-ty. He received his B.E. and M.E. in1995 and 1996, respectively, fromthe Department of InformationScience at Kyoto University. Hehas been working on a project

that aims to create systems that support a communi-ty of humans using a computer network. His e-mailaddress is yamaki@kuis.kyoto-u.ac.jp.

Takaaki Komura is currently agraduate student in the Depart-ment of Information Science atKyoto University. He received hisB.E. from Kyoto University in1997. He has engaged in researchin information-processing multi-cast technology and layeredencoding of multimedia data. His

e-mail address is emon@kuis.kyoto-u.ac.jp.

Nobuyasu Itoh is currently a student in the Gradu-ate School of Human and Environmental Studies atKyoto University. He received his B.E. from KyotoUniversity in 1997. He has been studying medicalinformatics and working as a member of the SensibleHuman Project. His research interests include apply-ing computer science to health care. His e-mailaddress is nobuyasu@kuis.kyoto-u.ac.jp.

Tadahiro Gotoh is currently aresearch scientist in the Commu-nications Research Laboratory,Ministry of Posts and Telecommu-nications, Kashima, Japan. He wasworking on the ICMAS’96 projectduring his stay at the Departmentof Information Science at KyotoUniversity in 1996. He has

engaged in research in time synchronizationthrough computer networks. His e-mail address istara@crl.go.jp.

Toyoaki Nishida is a professor in the Artificial Intel-ligence Laboratory at the Graduate School of Infor-mation Science at the Nara Institute of Science andTechnology. He received his B.E., M.E., and doctor ofengineering from Kyoto University in 1977, 1979,

tion for the Electronic Marketplace, White Paper1994, General Magic, Sunnyvale, California.

Yoshiyasu Nishibe is a researchengineer at NTT Information andCommunication Systems Labora-tories. He received his B.E. andM.E. in 1987 and 1989, respective-ly, from the Department of Electri-cal Engineering, Shinshu Universi-ty. His research interests includeagent-communication services,

multiagent system, communityware, and internetservices. His e-mail address is nishibe@isl.ntt.co.jp.

Hiroaki Waki is a research engi-neer at NTT Information andCommunication Systems Labora-tories. He received his B.E. andM.E. in 1989 and 1991, respective-ly, from the Department of Infor-mation and Computer Sciences,Toyohashi University of Technolo-gy. His research interests are agent-

communication services, mobile computing, andreal-world–oriented interfaces. His e-mail address iswaki@isl.ntt.co.jp.

Ichiro Morihara is a general man-ager of Kansai Systems Develop-ment Center at NTT KansaiRegional Headquarters. He re-ceived his B.E. and M.E. fromKyoto University in 1978 and1980, respectively. His researchinterests include knowledge engi-neering, agent-communication

systems, and mobile computing. His e-mail address ismorihara@sdc.bch.ksi.ntt.co.jp.

Fumio Hattori is the executivemanager of Computer ScienceLaboratory, NTT CommunicationScience Laboratories. He receivedhis B.E. and M.E. in 1973 and1975, respectively, from theDepartment of Electrical Engineer-ing, Waseda University. Hisresearch interests include intelli-

gent systems, software agents, multiagent systems,and communityware and social interaction. His e-mail address is hattori@eslab.kecl.ntt.co.jp.

Toru Ishida is currently a profes-sor in the Department of SocialInformatics at Kyoto University.He wrote Parallel, Distributed, andMultiagent Production Systems(Springer-Verlag), Real-Time Searchfor Learning Autonomous Agents(Kluwer), and Communityware:Concept and Practice (Wiley). His e-

mail address is ishida@kuis.kyoto-u.ac.jp.

Articles

48 AI MAGAZINE

and 1984, respectively. Hisresearch interests cover knowledgemedia, agent technology, knowl-edge sharing, and qualitative rea-soning. He is an area editor (intel-ligent systems) of New GenerationComputing, an editor of Auton-omous Agents and Multiagent Sys-tems, a fellow of the Foundation of

Physical Standardization Agents, and a trustee of theJapanese Association for Artificial Intelligence. His e-mail address is nishida@is.aist-nara.ac.jp.

Hideaki Takeda is an associateprofessor in the Artificial Intelli-gence Laboratory at the GraduateSchool of Information Science atthe Nara Institute of Science andTechnology. He received his Ph.D.in precision machinery engineer-ing from the University of Tokyoin 1991. He has conducted re-

search on a theory of intelligent computer-aideddesign systems, in particular experimental study andlogical formalization of engineering design. He isalso interested in multiagent architectures andontologies for knowledge base systems. His e-mailaddress is takeda@is.aist-nara.ac.jp.

Atsushi Sawada is an associate professor in the DataProcessing Center at Kyoto University. He receivedhis B.E., M.E., and doctor of engineering from KyotoUniversity in 1990, 1992, and 1997, respectively. Hisresearch interests cover software-engineering envi-ronments and information modeling and sharing forsoftware development. His e-mail address is sawa-da@kuis.kyoto-u.ac.jp.

Harumi Maeda is a JSPS ResearchFellow and a Ph.D candidate inthe Artificial Intelligence Labora-tory at the Graduate School ofInformation Science at the NaraInstitute of Science and Technolo-gy. She received a B.A. in psychol-ogy from Kyoto University in 1986and an M.Sc. in computation from

the University of Manchester Institute of Science andTechnology in 1994. She worked for Fujitsu Ltd. as asystems engineer from 1986 to 1995. Her researchinterests include knowledge representation, humanfactors in computation, and creative thinking sup-port. Her e-mail address is harumi-m@is.aist-nara.ac.jp.

Masao Kajihara is an M.E. studentin the Artificial Intelligence Labo-ratory at the Graduate School ofInformation Science at the NaraInstitute of Science and Technolo-gy. He received his B.E. in 1996from the Department of Informa-tion Science at Kyoto University.His research interests include aug-

mentation of human memory, knowledge represen-tation, and multiagent systems. His e-mail address ismasao-k@is.aist-nara.ac.jp.

Hidekazu Adachi is an M.E. stu-dent in the Artificial IntelligenceLaboratory at the Graduate Schoolof Information Science at the NaraInstitute of Science and Technolo-gy. He received his B.E. from theDepartment of Environmentaland Sanitary Engineering, KyotoUniversity, in 1996. His research

interests include ontologies, computer-aided design,and text analysis. His e-mail address is hideka-a@is.aist-nara.ac.jp.

Multiagent Systems Proceedings

ICMAS–96—Kyoto, Japan

Proceedings of the Second International Conference on Multiagent SystemsEdited by Edmund DurfeeThe proceedings covers a broad spectrum of perspectives, includ-ing communications issues and negotiation strategies.

ISBN 1-57735-013-8 476 pp., index, $59.95 softcover

Published by The AAAI Press

ICMAS–95—San Francisco, California

Proceedings of the First International Conference on Multiagent SystemsEdited by Victor LesserThe proceedings covers a broad spectrum of perspectives, includ-ing communications issues, and negotiation strategies.

ISBN 0-262-62102-9 472 pp., index, $70.00 softcover

Published by The AAAI Press / The MIT Press

For more information on these and other proceedings published by AAAI Press, visit our web site at

www.aaai.org/Press/

AAAI Press445 Burgess Drive

Menlo Park, CA 94025

Articles

SUMMER 1998 49