I-Views: a community-oriented system for sharing...

Computer Networks 33 (2000) 567–581www.elsevier.com/locate/comnet

I-Views: a community-oriented system for sharing streaming videoon the Internet

Pengkai Pan 1, Glorianna Davenport 1

Massachusetts Institute of Technology, Room E15-435 Media Laboratory, 20 Ames Street, Cambridge, MA 02139, USA

Abstract

Streaming media is pervasive on the Internet now and is continuing to grow rapidly. Most streaming media systems haveadopted the model of broadcast. Unfortunately, the nature of the broadcast-like one-to-many communication model is notable to foster interaction and collaboration among people. In this paper we discuss a community-oriented communicationmodel for sharing streaming video and introduce a prototype, I-Views. This is a system that permits individuals to usepublished, communally-owned media clips in order to author narratives by assembling clips, and to build communities ofsimilar interests based on comparison of these narratives. By offering shared authorship, tools and virtual environments,I-Views demonstrates some potential directions for ‘sharable video.’ 2000 Published by Elsevier Science B.V. Allrights reserved.

Keywords: Sharable video; Streaming media; Virtual community; Collaborative filtering; Retelling stories

1. Introduction

From the earliest civilization, news and storieshave been perpetually shared and retold. Recentstory technologies — movies, radio, TV — offera one-to-many channel for broadcast of stories butlack back channels for modification or discussionof these stories by the audience. The technologyseparates the teller from the mass audience. Lackingan obvious channel for interchange, it is very difficultfor participants to comment on news, present theirown interpretations of fiction, add their own material,or share their thoughts, experiences and impressionswith other audience members (Fig. 1).

The decentralized Internet infrastructure providesan efficient means for people to distribute and dis-cuss digital content. There are at least three ma-

1 E-mail: {Ppk, Gid}@media.mit.edu

jor reasons we believe that the adoption of shar-ing video-based content on the Internet will growrapidly. First, the infrastructures are developingrapidly. According to RealNetworks, more than 70million users have registered to use the RealPlayer[22]. By 2003, almost 26 million broadband sub-scribers will be connected to the Internet [9] andthese connections will ease access to video mate-rial. Second, the penetration of low-cost PC camerasand digital video cameras is increasing daily. IDC’sresearch [12] suggests that worldwide PC camerashipments will surge from 606,000 in 1997 to 9.2million in 2002, a compound annual growth rate of72.3%. These low-cost video cameras in combina-tion with powerful personal computers will facilitateindividual creation and publication of video stories.Third, people have used a variety of tools and sys-tems to share text-based information. Chat Room[15], ICQ [11], Newsgroup, and Third Voice [19]

1389-1286/00/$ – see front matter 2000 Published by Elsevier Science B.V. All rights reserved.PII: S 1 3 8 9 - 1 2 8 6 ( 0 0 ) 0 0 0 5 2 - 9

568 P. Pan, G. Davenport / Computer Networks 33 (2000) 567–581

Fig. 1.

have attracted multiple millions of users to exchangeinformation, impressions and feelings on a daily ba-sis. Applications in the domains of training, distancelearning, and entertainment also need to share video-based content on a daily basis. We imagine that, inthe very near future, millions of hours of streamingvideo will be shot by $200 digital video cameras,edited on 5000 MHz computers and distributed overgigabyte fiber Internet connections.

Further, broadband today is ill-suited to the re-quirements of either learning or entertainment. Mostbroadband programming is fragmented by advertis-ing and the full window interface of current cumber-some broadcast-like models [9] requires viewers toseek the narrow text-based channel of today’s Inter-net if they wish to share their impressions or ideas.New approaches to communication suggest that theintegration of interaction can help us create, learnand collaborate more effectively. Inspired by the‘intercreativity’ idea [1], a term coined by Tim Bern-ers-Lee, we have addressed the following questions.What type of interchange might be developed to en-courage people to share streaming video on the Inter-net? Can we create easy-to-use models that facilitatesharing and reusing available content which exists orwill exist on the Internet? What are the best mech-anisms for people to initiate conversations aroundstreaming video-based content? How can video fos-ter virtual communities or sub-communities?

In the past two years, we have focused on design-ing a system for asynchronous sharing of stream-ing video on Internet-like distributed networks. Wehave built a prototype, called I-Views, which allowswidely distributed groups of people to view, edit,compare, evaluate, and discuss streaming video ma-terial over the Internet. I-Views presents the userwith two types of tools: Web-based video editingtools and virtual community-building tools. The for-mer allows the user to view, select, save, re-sequenceand publish video clips; the latter allows the user toinitiate dialogues by matching common interests andassumptions and to build virtual communities aroundstories. In marrying these two types of communica-tion — streaming video-based viewing=editing toolsand email tools — audience members are able to cre-ate and share content. By creating active dialoguesaround reconfigured sequences, the audience canenjoy an entertaining and intellectual interchange(Fig. 2).

Currently, we are evaluating the prototype systemof I-Views using video footage documenting the Ju-nior Summit ’98 [18], which was a cross-cultural,cross-geographical and multi-language project in-volving thousands of children, Junior Summit staff,volunteers, sponsors, and MIT Media Lab facultyand students. The content reflects children’s think-

Fig. 2.

P. Pan, G. Davenport / Computer Networks 33 (2000) 567–581 569

ing about world culture, child labor, kids banks,telecommunications access and environmental re-sponsibility. Specifically, an international group offilmmakers including four junior filmmakers havecontributed more than one hundred hours of footage.The current video database includes eighty stream-ing video clips selected and edited from the footageshot. We will use this particular example throughoutthis paper.

The paper is organized as follows. We first presenta context section in which we discuss related workand our approach. We then describe the structure andimplementation of I-Views, its client=server archi-tecture and the tools developed for shareable video.The third section presents a preliminary evaluation.Finally, we draw conclusions about our contributionsand offer future suggestions.

2. Context

In this section, we basically discuss two ques-tions: what the participant is able to share and withwhom s=he can share. We also discuss researchworks related to our studies.

2.1. What to share

The concept of ‘Sharable Video’ should not belimited to sharing video content only. We need torethink ‘Sharable Authorship’ and ‘Sharable Tools’on the Internet. A decade ago, Professor GloriannaDavenport developed a ‘Sharable Video’ project,New Orleans in Transition [6], on a UNIX work-station supported by a bank of video disc players.The purpose of the project was to give students ofUrban Planning an insider view into the process ofchange. Students using this material could make useof an editing tool which allowed them to excerptfrom the available segments, re-sequence and com-pile movies. These movies could then be imbeddedinto the systems which they developed for the class.On using the systems, first without the video andthen online with the video, Professor Denis French-man at MIT commented that the video clips addedenormously to the overall argument of the systems.Nowadays, millions of families and friends sharea variety of still pictures and video. Recycling or

sharing other peoples’ work is more efficient thaneach person repeating similar work. For example,one person has shot a beautiful landscape of Parisfrom an eagle-eyed perspective. His or her friendstraveling to Paris would like to reuse the same shotand make their own travel stories about Paris. Asmentioned above, ubiquitous digital cameras and fastInternet connections will definitely produce millionsof hours of video content on the Internet. Sharingvideo content as well as sharing authorship will bewidely accepted by mutual consent.

Some related work has been developed in theWeb-based editing field from varying perspectives.In VideoFarm.com [20], users can download aJava-based Premiere-like editing software, edit theirfootage, then upload final cuts to VideoFarm’s server.VideoFarm [20] is a combination of a Web-basedediting application and video content hosting ser-vice. VideoFarm is developed mainly from the con-tent author’s point of view. Mimicry [2] is a systemallowing users to create links to and from temporal(video and audio) media on the Web regardless ofthe users’ ownership of the Web pages or mediaclips involved. Mimicry demonstrates additional fea-tures for current media players. Unlike VideoFarm,Mimicry is created mainly from the audience’s pointof view.

In I-Views, we rethink the roles of the author andthe audience in sharable media environments on theInternet. In these environments, the distinction be-tween the author and the audience becomes blurred.All participants are able to share authorship, videocontent and on-line tools. The participants can saveand edit their favorite video clips and publish theirnew sequences back to I-Views without applying forpermission from the authors of the original clips.The sharable authorship offers each individual free-dom to retell stories from different perspectives andassumptions. On the other hand, all retold sequencesrecord the original information about the sequences,such as who the authors of the original clips are,when the clips were made, etc.

2.2. Who will share with whom?

In her doctoral dissertation, Professor Judith S.Donath comments “The problem [has been] designsfor data, not people.” She stated how “visitors to a


site are likely to share common interests, yet theycannot communicate with each other, nor are theyaware of each other’s presence” [7]. The most criticalmechanism lacking in current networked storytellingsystems is the mechanism of initiating conversationsaround streaming video-based content such that itfosters the construction of virtual communities. Avery real question we must ask in constructing anenvironment for sharing video is: Who will sharevideo with whom?

The Junior Summit ’98 generated an on-line con-versation of over 2000 children; one hundred ofthose children later convened at the MIT Media Lab.The first night these children were given pagers forreal-time messaging. Different people have differ-ent skills, interests and passions. Some children atthe Summit loved playing LEGO while others fo-cused on children’s rights. Building matchmaker-likeagents for their real time communications (see previ-ous case study of ICQ) [16] was a useful approach tobringing individuals with similar interests together.In our first case-study of I-Views, we built a databaseof over 100 short movies representing the diverseactivities of the Junior Summit. The children of theSummit were our test users. Having been at the Sum-mit, they had an interest in the video representationof the event. Would they edit and message with it?

The question of “who will share with whom?” isinformed by examples in the growth of various virtualcommunities, as well as by research in collaborativefiltering techniques. Pascal Chesnais’ Canard [4] the-sis takes a constructionist approach and outlines theASE framework, which represents the three attributeswhich contribute to usability within a shared tech-nological environment: Ability, Support and Effort.Amy Bruckman’s MOOSE Crossing [3] encouragesinteractive chats through the use of ‘rooms’ and ‘ob-jects’ for building communities. Rooms are similarto chat channels and objects are artifacts transportedfrom one room to others. Both studies are inspiringin terms of understanding how virtual communitiesemerge and grow. In I-Views, the objects are the videomaterials and the email messages.

Once we have a database of users, how do theyfind each other? The basic goal of collaborative fil-tering is to cluster people who have similar interestsin order to grow the community itself. In I-Views, themechanism of the comparison algorithms is able to

analyze the sequence that is made by the participantand compare that sequence to all other sequences inthe database. I-Views then presents the ten most sim-ilar sequences to the participant. A user might watchsome of the ten sequences and decide whether or notto initiate conversations. The comparison tool is ameans for participants to match their common inter-ests and foster new communications with others. Thecomparison algorithms look similar to prior collab-orative filtering methods [8,13,14]. However, thereare a few significant differences in terms of user dataretrieval, objectives, and procedures. We compareour algorithms to that used in Ringo [14]. Ringo isan on-line music recommendation system which hasmechanisms to recycle word-of-mouth knowledge bymatching common interests among users.

2.2.1. Data retrievalIn Ringo, users have to vote which song is good

or bad. The processes to retrieve users’ feedback arevery time-consuming and subjective. I-Views takesa different approach. We assume that the clips thatthe user selects make sense to that user. Participantsneed not vote. The process of retrieving participants’feedback is transparent in I-Views.

2.2.2. ObjectivesThe objective of Ringo is to make personalized

recommendations. The users expect to get accuraterecommendation results. However, because of thenature of the algorithm, the method cannot workwell if the user base is not large enough. In I-Views,the comparison tools offers a means for participantsto find other people with whom to initiate dialogues.Our focus is on matching people. The comparisontools also work even when there are very few partic-ipants using the system.

2.2.3. ProceduresIn Ringo, the procedure of recommendation is

static and black-boxed. The user votes on a set ofsongs, then Ringo gives out a list. I-Views’ com-parison tools are dynamic. The participant can movethe scroll bars to justify the criteria and define thesimilarities. The movements of thumbnails react tothe changes of scroll bars simultaneously. The partic-ipant is able to perceive the interconnection amongthese sequences. We are still testing and improving


our comparison algorithms and interface design. Webelieve that meaningful comparison tools can bringdistributed participants closer together.

Other related work of interest includes XMLand SMIL. While World Wide Web Consortiumis rapidly developing new languages and standards[21], SMIL 2.0 [17] is supposed to have more com-pelling features than the 1.0 version [10]. Unfortu-nately, few features have been proposed which willfacilitate collaboration. Collaboration might involveone video editor working with two composers ona soundtrack for a movie which will be shown onthe Internet. The new version of SMIL should en-able multiple audio tracks to be shared, synchedand unsynched in a distributed exchange such thatthe editor can listen to a mixed track incorporatingchanges from both composers even while comment-ing on the effect. We would like to see that WorldWide Web Consortium pay increased attention tothe user’s point of view which is different from thatof the broadcaster. It is useful to remember that ininteractive expression, a single person will alternatebetween the role of the author and the role of theaudience.

3. Implementation

I-Views has evolved monthly. In this chapter, wepresent the version of July 1999: client and serverside modules, database modules, net work architec-ture and data flow. We also discuss the comparisonalgorithms.

3.1. Modules on client side

There are seven major components on client side:(1) User registration and login;(2) Viewing original video clips;(3) Editing video clips;(4) Comparing video sequences;(5) Finding the most popular sequences;(6) Contributing video content;(7) Searching for content.

In this paper, we discuss the following three com-ponents: viewing original video clips, editing videoclips and comparing video sequences. For the othercomponents, please read Pengkai’s Master’s thesis.

3.2. Viewing original video clips

In our test application, approximately eightythumbnails are loaded into the Viewing applet(Fig. 3). Each thumbnail is linked to a video clipin the video database. The participant is able to movethe yellow oval scroll bar to browse the images.The sizes and positions of the images change basedon the movement of the scroll bar. In our first im-plementation, the presentation of images is random;ideally, the scroll bar would allow the user to browsethe material according to certain criteria, such aschronology, subjects and characters. The participantcan double-click on any thumbnail, and a Real G2player pops up and plays the associated video clip(Fig. 4). If the participant finds any of the clips in-teresting, s=he can single-click on the clips and savethem into her or his personal archive. The selectedimages have a thin yellow outline around them forease of selection. After the participant selects his orher favorite images, s=he can click on the Next Pagebutton and go to the editing page.

3.3. Editing video clips

This applet is a simple on-line, real time sequenc-ing tool (Fig. 5), which allows the participant to dragand drop images into the editing box to sequencevideo clips into a longer assembly. The participantcan modify this sequence by inserting, deleting, andswitching clips. To preview the sequence, s=he clickson the preview button and the G2 player pops up.When the participant is satisfied with the sequence,the participant gives the sequence a name and broad-casts it. To offer the on-line real time sequencingtool, we adopted SMIL language. The language isable to precisely synchronize the timing of diversemedia elements as well as the presentation of theseelements through a SMIL-enabled player such as aG2 player, which is also used in our environment.When the participant links some clips to make a se-quence, the Java server and Perl scripts automaticallygenerate SMIL tags and call the G2 player. The G2player is able to play these clips seamlessly one byone according to the tags. Because of the nature ofSMIL language, we will be able to add more featuressuch as fade in=out and trimming clips later. Thesimplicity is the beauty of this tool.


Fig. 3.

Fig. 4.


Fig. 5.

Fig. 6.


Fig. 7.

3.4. Comparing video sequences

After the participant broadcasts the sequences, heor she is able to view the sequence by clicking on theSTEP TWO button. There are three columns on thispage: Snapshot, Sequence Name, and Comparison(Fig. 6). To view the sequence that the participantproduced, the viewer clicks on the associated snap-shot and a G2 player plays the video clip. To findout more information about the sequence, the viewerclicks on the name of the sequence and a new win-dow pops up (Fig. 7). The window presents therelated information: the author, the clips contained inthe sequence, the date the sequence was produced,etc. To share the sequence with friends, the viewerclicks on the Go button and a message window popsout. The viewer can send out an email to his orher friends, who will receive a URL link to thatsequence. By clicking on the URL, the receiversare able to view the sequence via a G2 player. Weprovide a comparison tool for the participant to findother participants who have similar interests. Click-ing on the Compare button, the participant is able to

explore using a radar-like interface. We have adoptedradar as a metaphor in interface design to providea useful means for very distributed participants tointeract.

In the middle of the applet, there is the sequencethat the participant selects as a metric for comparisonwith other sequences in the database (Fig. 8). Aroundthe selected sequence, the ten most similar sequencesare displayed. On the right side of the applet, thereare four criteria: Clips, Keywords, The Beginning,and The Ending. The criteria allow the participantto determine what it means to be similar; similarityis communicated in a display where thumbnails ofthe movies appear nearer to or farther from the cen-ter. The distance between the selected sequence andanother sequence suggests how similar the two se-quences are based on different criteria. For example,if the participant wants to find out which sequenceshave the same beginning as the selected sequence, heor she can move the scroll bar of The Beginning tothe position of one hundred percent (Fig. 9). Threesequences that have the same beginning move closerto the center. Or, the participant can use a combina-


Fig. 8.

tion of the four criteria: for example, the scroll bar ofKeywords can be positioned to fifty percent and thescroll bar of The Beginning to one hundred percent.Or, by setting all scroll bars to one hundred percent,the participant can easily see the movies which aremost different, hence farthest away. As the partici-pant discovers sequences that s=he is interested in,s=he clicks on the sequence to learn the author andtitle of the sequence. In the example (Fig. 10), the au-thor is Vincent and the title is Interview with Nic. If aparticipant wants to initiate a dialogue with Vincent,s=he can click on the Go button to get a messagewindow from which to send a message to Vincent.So far, email is the only means that we provide forparticipants to initiate dialogues. We are consideringof designing a synchronous communication systemin future versions of I-Views.

3.5. Comparison algorithms

Here, we briefly introduce the algorithms thatwere used to implement the comparison scale in

I-Views. The comparison scale helps the participantfind out which sequences are similar to the sequencethat they selected as a metric. Two factors determinehow similar two sequences are: the subjective factorand the objective factor. The subjective factor isdetermined by the percentage of a criterion; theobjective factor is determined by the number ofcommon clips or keywords between two sequences.The high level algorithm is as follows.

W Dn�1XjD0

��1

n

�Ł .Sf j Ł Of j /

�where W is the weight for measuring the similaritybetween two sequences; n is the number of crite-ria (in I-Views’ case, there are four criteria;) Sf isthe percentage of the subjective factor; Of is thepercentage of the objective factor.

For example, say we have a metric Sequence Awhich includes Clip 1, Clip 2 and Clip 3 and twoother Sequences to be compared with it: SequenceB, including Clip 1, Clip 2 and Clip 5, and Sequence


Fig. 9.

Fig. 10.


C, including Clip 3, Clip 4, and Clip 6. Let’s saythat we only want to make the comparisons basedon similar clips between two sequences. So we movethe scroll bar of the Clips to one hundred percent. Ofis a variable determined by the number of commonclips between two sequences.

Of D the number of common clips

the number of clips in the longer sequence:

The number of Sequence B’s Of is: Of D2=Max.3; 3/ D 0:667, while the number of SequenceC’s Of is: Of D 1=Max.3; 3/ D 0:333. Now, n is 4and Sf is 100%.

The weight of Sequence B is:

W D . 14 / Ł .1 Ł 0:667/ D 0:1667:

The weight of Sequence C is:

W D . 14 / Ł .1 Ł 0:333/ D 0:0833:

The weight of Sequence C is smaller than theweight of Sequence B, since Sequence A and Bshare two of the same clips while Sequence A and Cshare only one. The keyword-based algorithm is thesame as the one described above, while the processof matching a similar beginning is similar to the oneabove.

Fig. 11.

3.6. Modules on server side

I-Views is a multimedia-based multi-client=serversystem built on various computer languages and us-ing a variety of tools. We took a client-orientedapproach in the design of the server side modules[5]. The design and implementation of server sidemodules are primarily based on the client side needs.We also considered the performance of communi-cation and computation as certain limiting hardwareand software resources exist. The whole system isbuilt on freeware, such as Linux, Java, Perl, Apache,MySQL, etc. To present the system’s back-end, wediscuss the system architecture, then briefly describePerl script modules.

All modules and data exist on three machines:wwwic.media.mit.edu, fellini.media.mit.edu and Yi-mou.media.mit.edu. On the three machines, there arefour servers running simultaneously: Web Server,Java Server, Database Server and the Real VideoServer (Fig. 11). The diagram shows the locationsof all servers, contents, applet classes, Perl scripts,etc. Generally speaking, the IC Web server is incharge of static HTML files and contains a mirrorcopy of the thumbnail images; Fellini contains Java


Fig. 12.

Web server, Servlets, Applets, JDBC driver classes,Perl programs, MySQL Database, etc.; Yimou is thestreaming video powerhouse.

Fig. 12 represents the Applet-Servlet modules forI-Views. On the left of the diagram, there is a JavaConstants class, which contains all of the URLs andpathnames in Servlet codes. In the middle of the dia-gram, there are eight independent Servlet and Appletmodules, which have different functions. Registra-tion Servlet handles new participants and transferstheir registration information to the database. Lo-gin Servlet deals with the participant who alreadyhas registered. If the participant logs in successfully,their browser is given a cookie which includes theparticipant’s identification number. The View Appletpackage contains the View applets that allow the par-ticipant to browse and watch the original video clips.The applets in the Edit Applet package deal withediting, previewing and broadcasting events. Near-est Servlet automatically generates the appropriateHTML tags, which have the Comparison appletsin the Nearest Applet Package. The TopTen Appletpackage does the computation to find the most pop-ular sequences based on hit rates. On the right ofthis diagram are the database related modules. Allapplet codes are associated with the JDBC driver:MySQL Driver Package. The SQLConnect Packageworks as an interface to the MySQL Driver Package.In I-Views, Perl scripts perform two types of jobs:generating the SMIL tags and processing support-

ing work such as searching keywords and sendingemail.

4. Evaluation

In our preliminary evaluation we engaged twosmall groups of users, a group of eight Junior Sum-mit delegates who used the system remotely anda group of Media Lab students and faculty. Ourevaluation focused on two objectives:(1) Will users, who are portrayed in a particular

set of observational video content, engage withthat content to the point of creating their ownvideo sequences from the set, publishing theirsequences to their friends and developing discus-sion communities around this material?

(2) Can we improve the workflow and interface de-sign of I-Views? Should aspects of I-Views beredesigned to better accommodate the commu-nity building objective?

4.1. The evaluation participants and methods

Group A: Our most interesting evaluation drawson observations about how a group of eight JuniorSummit delegates used the system remotely. Thisuser group was critical to evaluating our main hy-pothesis, that people presented with a system inwhich they can share, reedit and publish video will


Fig. 13.

do so. All members of this group had participated inthe six-month on-line forum and the one-week Me-dia Lab Summit. They knew each other and workedtogether through a variety of projects, discussiongroups and email lists. From June 12, 1999 to July12, 1999, they volunteered to test I-Views. Eight ofthem successfully registered in I-Views and six ofthem were able to view video clips from I-Views.Four participants retold stories and two of them sentout email and invited other people to join in I-Views(Fig. 13). We discuss the details later.

Group B: Our second group of users, while notengaged in community building, were selected tocomment on the actual design of the systems atthe MIT Media Lab. During the on-site evaluationprocesses at Media Lab, six participants were ableto register in I-Views and watch video clips. Four ofthem tried to retell stories. Three of them tried theemail function; none of them invited other people towatch or discuss their reedited segments (Fig. 14).

Following the hands on portion of the evaluation,we talked to participants to understand more fullytheir engagement with the system. Our evaluationresults relate to three topics: accessibility, workflowand interface design and virtual community building.

4.1.1. AccessibilityTo engage with I-Views, the participant needs a

fast Internet connection such as a cable modem, a

Fig. 14.

DSL, ISDN or T1. The cost of Internet connectionand low data rates are two major reasons why theycouldn’t use I-Views. For example, one participantsaid:

Sorry but I don’t think I’ll be able to use I-Viewsas I’m not allowed to pay in order to download theG2 real player. I wish I could help. (Konstantina)The bandwidth is a big barrier for most people

experimenting with I-Views.

4.1.2. Workflow and interface designWe have received many valuable comments on

workflow, function design and interface design. Mostpeople like the interface design and explicit work-flow design. However, some users were confused bysome parts of I-Views, particularly, the View appletand the comparison tool.

Maybe because I was using a portable, but manywere very small I’d no sooner locate one and viewit and mark it then I’d lose the place where thenext one for my sequence was. It was a lot likethose memory games where you have to rememberwhere the face down cards are in order to turn apair over: : : As I said before, I didn’t like thosepictures moving around — kind of neat in the be-ginning to watch but very frustrating to work with.I would have preferred something more static andlarger. I think that I would have then made manymore sequences and sent them to people. (Vincent,age 13, Paris)Children of different ages respond quite differ-

ently to the comparison module. Children from 10to 12 years old don’t understand the comparison toolat all. We assume the reason is that the design ofthe tool is not clear and compelling enough. How-ever, we found that the older children from 13 to 16years are more willing to try the comparison tool,retell stories and send out messages to their friendsand family than the younger children are. They alsoenjoy receiving the comments that their friends sendback.

4.1.3. Community buildingWe have not had a large enough group using

I-Views to comment on its ability to build com-munity. In the Junior Summit database, participantsretold stories focused on only two topics: LEGO andthe international fashion show. While they did share


their stories among themselves and their friendsthrough email, the system did not encourage themto submit new materials, nor would many of themhave had the means to shoot, edit and upload video.Obviously creating a system that grows communityrequires the right mix of content invitation and tools.Neither is sufficient on its own.

As one reader of this paper commented “asidefrom having discussions about the finished product,get users to collaborate and=or help each other duringthe production process. For example, set up a com-petition. Divide people into groups and have themcollaborate for the competition.” Competition in con-junction with a constructive environment represents acritical aspect of World Wide Web initial popularity.

Within our small user group, we observed thatdifferent participants assumed different roles. Threebasic roles are: observer, builder and leader. Amongeight testers, about four participants were observers,two of them were builders and two of them wereleaders. We also had a few new users who were in-vited by the two leaders. These leaders played a keyrole in terms of self-organizing virtual community.For example, Vincent not only watched a lot of videoclips, retold stories with extreme patience, but healso sent sequences to his moderator, Julia, and in-vited her to join the test of I-Views. When Julia hadproblems with using I-Views, he immediately sentemail to us and tried to solve her problems. Sincethe leaders played extremely active roles in growinga community, we will in the next redesign focus onfunctionality to promote the leader role for buildingup and organizing sub-communities.

5. Conclusion

In this paper we have examined a hypothesis:Can we design and implement a distributed systemfor video that encourages the social use of backchannel to build a sense of community? Inspiredby Tim Berners-Lee’s ‘intercreativity,’ we have builtI-Views, a prototype of Sharable Video that presentsthe stories of a global event, the Junior Summit ’98.Specifically, there are two major contributions of thisaspect of the work:(1) A democratic sharable storytelling model which

enables very distributed participants to share au-

thorship, tools, spaces, impressions and experi-ences around video stories on the Internet. Sto-ries about the Junior Summit ’98 were used asour video database as the children and parentswho participated in this event offered an existingcommunity of interest.

(2) A dynamic similarity scale: this tool attempts tomatch common interest through various metricsof similarity. By identifying common interest,such a tool can bring participants together whomight like to engage in conversation around atopic. In turn, this could encourage the social useof back channel to build the sense of community.

Our hypothesis has been explored through the im-plementation of I-Views and on-site and remote testsand evaluation. This Sharable Video environment in-vites participants to use the back channel to builda community of interest. In our limited case study,we observe three types of participants: Observer,Builder and Leader. These types may prove usefulin the construction of a tiered application in whichcomplexity is matched to the type of user.

Many thanks to our undergraduate research as-sistants, Christina Chu, Vikas Sodhani, Alice Yangand Carmen Woo. Their creativity and diligent workbuilt the working system, I-Views. This researchwould not have progressed as far and as quicklywithout them. We thank Prof. Walter Bender, Prof.Brian Smith, and Prof. Bruce Blumberg for excellentfeedback and advice. Thanks to our Interactive Cin-ema colleagues: Dr. Kevin Brooks, Brian Bradley,Barbara Barry, Ricardo Torres, Paul Nemirovsky,Aisling Kelliher, James Seo, Roger Sipitakiat, ArjanSchutte and Phillip Tiongson for kindness and sup-port. This work is supported in part by a grant fromthe News in the Future and Digital Life Consortia.

References

[1] T. Berners-Lee, Focus on intercreativity instead of interac-tivity: create things together, in: Plenary Sessions, 5th Int.World Wide Web Conf., May 1996, Paris, France.

[2] N.O. Bouvin and R. Schade, Integrating temporal mediaand open hypermedia on the World Wide Web, in: 8th Int.World Wide Web Conf., Toronto, Canada, May 1999.

[3] A.S. Bruckman, MOOSE crossing: construction, commu-nity, and learning in a networked virtual world for kids,Ph.D. Thesis, MIT Media Arts and Sciences, June 1997.


[4] P.R. Chesnais, A framework for designing constructionistapproaches to community-centered messaging, Ph.D. The-sis, MIT Media Arts and Sciences, 1999.

[5] C. Chu, Technical architecture and implementation ofI-Views: an online storytelling community, Advanced Un-dergraduate Project Final Report, 1999.

[6] G. Davenport, New Orleans in transition, 1983–1987: theinteractive delivery of a cinematic case study, Int. Congressfor Design and Planning Theory, Education Group Conf.Proc., May 1987.

[7] J. Donath, Inhabiting the virtual city: the design of so-cial environments for electronic communities, Ph.D. Thesis,MIT Media Arts and Sciences, 1997, http://judith.www.media.mit.edu/Thesis/.

[8] D. Goldberg, D. Nichols, B. Oki and T. Douglas, Usingcollaborative filtering to weave an information tapestry,Commun. ACM, December 1992.

[9] M.E. Hardie, Hooked on broadband, The Forrester Report,July 1999.

[10] P. Hoschka (Ed.), Synchronized multimedia integration lan-guage, World Wide Web Consortium Recommendation,June 1998, http://www.w3.org/TR/REC-smil/.

[11] ICQ, A Net-wide instant messenger, http://www.icq.com/icqhomepage.html.

[12] IDC Research, IDC research brings PC camera marketinto focus – shipments to exceed 9 million in 2002, Feb.1999, http://www.idg.com/cgi-bin/pressroom.cgi?566E3ADB8F45B1A18525672900772037.

[13] J. Konstan, B. Miller, D. Maltz, J. Herlocker, L. Gordanand J. Riedl, GroupLens: applying collaborative filtering toUsenet news, Commun. ACM, March 1997.

[14] P. Maes and U. Shardanand, Social information filtering:algorithms for automating ‘Word of Mouth’, CHI, 1995.

[15] J. Oikarinen and D. Reed, Internet relay chat protocol, Net-work Working Group Request for Comments (RFC1495),May 1993.

[16] P. Pan, I-Views, a storymaking community of, by and forthe audience, P22, Master Thesis, MIT Media Arts andSciences, 1999.

[17] L. Rutledge, J. van Ossenbruggen and L. Hardman, Antici-pating SMIL 2.0: the developing cooperative infrastructurefor multimedia on the Web, 8th Int. World Wide Web Conf.,Toronto, Canada, May 1999.

[18] The Junior Summit ’98, http://www.jrsummit.net.[19] Third Voice, Third Voice helps people post public or per-

sonal notes on any Web Page, http://www.thirdvoice.com/products/product_features.htm.

[20] VideoFarm, http://www.videofarm.com/.[21] XML Extensible Markup Language, http://www.w3.org/X

ML/.[22] Yahoo Finance, http://biz.yahoo.com/e/l/r/rnwk.html

Pengkai Pan is a Ph.D. candidateand works as a research assistant forthe Interactive Cinema group at theMIT Media Lab. He earned a B.E. inindustrial design and a minor degreein Computer Science with highesthonors from ZheJiang University in1994. He then joined the Departmentof Electrical Engineering and Com-puter Science as an assistant profes-sor. In 1999 he earned a Master’sdegree from MIT Media Lab, where

he designed I-View, an Internet-based, streaming video, story-sharing system. He continues to investigate ‘sharable media’,with the goal of inventing efficient and novel means for sharingvisual information. He is an Interval Fellow.

Glorianna Davenport is the headof the Interactive Cinema group atthe MIT Media Laboratory. Between1988 and 1995, Davenport’s grad-uate workshop in multimedia pro-duction was adopted by five interna-tional institutions. Her publicationson subjects of responsive media, aswell as several of her interactive dig-ital prototype programs, have beenincluded in international symposia,conferences, and film festivals. In-

formation on some of these projects is accessible on the Web(http://ic.www.media.mit.edu/). She serves on the editorial boardof IEEE Multimedia, and on the advisory board of several dig-ital content companies and cultural organizations. Trained as adocumentary filmmaker, she has achieved international recogni-tion for her work in new media forms, content, and deliverysystems. Her research explores fundamental issues related to thecollaborative co-construction of digital media experiences, wherethe task of narration is split among authors, consumers, andcomputer mediators. Ms. Davenport’s recent work focuses oncustomizable, personalizable storyteller systems that dynamicallyserve and adapt to a widely dispersed society of audience and onmobile media systems.

I-Views: a community-oriented system for sharing...

Documents

Transcript of I-Views: a community-oriented system for sharing...