Scalable Collaborative Virtual EnvironmentConsidering User’s Interest Information

8/3/2019 Scalable Collaborative Virtual EnvironmentConsidering User’s Interest Information

http://slidepdf.com/reader/full/scalable-collaborative-virtual-environmentconsidering-users-interest-information 1/5

Scalable Collaborative Virtual Environment

Considering User’s Interest Information

Yushi Nakai

Graduate School of Software and

Information Science

Iwate Prefectural University

[email protected]

Naoki Kamon

Faculty of Software and

Information Science


[email protected]

Yoshitaka Shibata


[email protected]

Abstract

Recently researches concerned with various applica-tions based on Collaborative Virtual Environment (CVE),such as virtual society, virtual shopping mall, virtual mu-seum etc., methods to realize cooperative works by usersin 3D virtual space and scalability issues of the VR space

to support a number of users have been investigated. How-ever, as the number of the users increases in CVE, usabil-ity of co-operative works by users degreases and scalableCVE cannot realized by using conventional client/server model. Also, in those researches, user interests the 3D objects in CVE are not considered.

In order to overcome the difficulties, we introduce a new method which uses P2P network to realize scalable CVE and recommendation system which uses user interests topresents and offer the proper objects in CVE. Using our suggest method, the objects can be distributed on the P2P networks and their computing loads can be reduced, even-tually the scalabilityof the CVEcan beimproved. Further-more, by reflecting user interest information to the appli-cation system, user can access and communication to 3D objects and contents which are more interested than theothers with higher priority. Thus total usability of the sys-tem can be improved. In this paper, we provide a new sys-tem and its architecture, P2P network to realize overlay network and methodology to realize user recommendation system to reflect user interest.

1. Introduction

Recently, as adventof broadbandnetwork technology, high-speed and regularly connectable network services have beendaily used. As a result, in addition to the telecommunicationusingtextand imagein thepastyears,more interactive andre-

altime communicationusing audio and video is beingexpectedin the recent years.Ontheotherhand,asprogressofcomputingpowerandmul-

timedia technology, application systems using high quality 3Dgraphics are well developed. As a result, a large scale collab-orative virtual environments (CVE) to support collaborativework by many users over high-speed network have been pro-posed so far [1] [5][7][8]. However, it is required to guaranteescalability and usability of CBV system and usability the en-hanced CVE communication function as the number of par-

ticipants increases from more several hundreds to thousandpeople using high quality video and audio communication fa-cility.

In the previous research, we investigated a CVE systemwhich can reflect both user interest information and user lo-cation information in 3D virtual space to resolve the scalabil-ity problem [8]. However, since in the previous research, theuser interest information had to be manually input to the sys-tem and updated once the user interest changed. For this rea-son, the system behavior and effect depended on the users in-putand update. Alsouser interestreflected onlyto theinterac-tion betweenusersbutnot tothe allof theobjects andcontentsin the3D virtual space. Therefore, byexpanding theuser inter-estto allof theobjects andcontentsin the3D virtual space,us-abilityby usercan be considered to be improved. Fromthis rea-son, inths paper,we proposea newCVEsystem whichis basedon a P2P network as infrastructure and user interest informa-tion and model for user model and control method to presentobjects and contents in 3D virtual space.

Inthe followings,the related worksis discussedin section 2.System configuration of suggestedCVE is explainedin section3. Suitable overlay network for CVE based on 2P2 networks isintroduced in section 4. Contents control layer for Level of De-

tail (LOD) is explained in section 5. Model of user interest toreflectto the3D object andcontents andis preciselydiscussedin section 6. The prototype system is briefly explained in sec-tion 7. Section 8 concludes our research.

2. Related Work

There are severalrelatedresearches concernedwith realiza-tion of CVE. In our previous research[8], we developed flexi-ble CVE system which realizes the following functions includ-ing;

• user retrieval function using user’s interest information,

• media selection function using interest, position and dis-tance of users and

• QoS priority and level control function using interest, po-

sition and distance of users.

By introducing those functions, network and computing re-sources could be distributedwith a high priority to theimpor-tant communication performed in 3D virtual space and QoScontrol of the objects and contents in 3D could be performedby user’s demand.

As related works, communication framework in 3D virtualspace [5] [1] [7] was investigated. In this framework, althoughthenetwork traffic is reduced bycontrolling QoSof the objects

Proceedings of the 20th International Conference on Advanced Information Networking and Applications (AINA’06)

550-445X/06 $20.00 © 2006 IEEE



based on their positions in the virtual space, the user interestcould not be taken into consideration.

In the research [4], a virtual conference system by CVE inwhich the user interest is modeled and the required informa-tion with objects was efficiently collected was proposed. How-ever, in this system, the interaction between users could notbe considered, but only interactionbetweena user andthe ob-

jects. Nor history of interest by of the user action was consid-ered.

In the research [6], recommendation function based on theuserinterest is introduced. Theresultof recommendationfromthe user action and the relations to the other user is presentedto the user. However, this method is not used for CVE sys-tem.

For those researches, in this paper,recommendationis pro-posed to use for collaborative work by providing the moreproperobjectsand contents in3D with high quality while scal-able CVEsystem canbe realize even thoughthe number of theusers and objects in 3D virtual space increases.

3. System Configuration

In this section, a new system architecture to support the

user interaction by recognizing the essential user’ requirementon the scalable CVE is proposed. Using this system, the usercan efficiently find and interact to the objects and contents in3D virtual space according to the degree of the interest. Thesystem architecture of our suggested CVE is shown in Figure1. and organized by 4 layers including Interface Layer, Inter-est Control Layer, Contents Control Layer and Overlay Net-work Layer. OverlayNetwork Layer, furthermore, dividedintoNegotiation Plane and Session Plane.

Interface Layer performs user interface functions includinguser’s operations, manipulations and presentations of 3D ob-

jects and contents.

Interest Control Layer performs analysis and recognitionfunctions of user interest by user’s action history and the re-lation to other users and control function of the 3D contents

to Interface Layer. By this function, the quality of the objec-tive 3D contents to be presented is properly adjusted.

Contents Control Layer performs objective 3D contentsmanagement to be controlled. The objective 3D contents inthissystemare classifiedas shown in Figure 2. Inthis Layer,the3D contents required from Interest Control Layer are foundedfrom CVE and provided with the proper quality. Also new 3Dcontents areadded andallocatedto CVE andmanaged in thislayer.

Overlay Network Layer provides logical network to real-ize CVE. Overlay Network Layer also includes search func-tion of 3D contents to respond to the user’s request and ex-change function of interest information between users. Nego-tiation plane performs session management functions for userto join to and release from the physical network. This planealso includes search function of node in overlay network to re-

spond to the request form the upper layer. Session Plane per-forms communication function to send andreceive interestin-formation and contents information after the session has b eenestablished.

4. Overlay Network Layer

In our system, Peer-to-Peer network (P2P) is ap-plied for Overlay Network Layer to organize an infrastruc-ture of CVE by interconnecting the user’s computes. Al-though client/server model or hybrid model are considered as

Figure 1. System Architecture

Figure 2. Classification of Contents

a possible infrastructure [3] [1], those models are disadvan-tage in that thetrafficand workloads concentrate to the serverand eventually lead to a bottleneck as the number of partici-pating users increases. On the other hand, the P2P networkalso hasa disadvantage inthat search andcollection of 3D con-tents are generally difficult although those contents can bewidely distributed over P2P network. In our suggested sys-tem, by applying for P2P network to eliminate the bottleneckinherent with conventional client/server model, and by tak-ing the characteristics of CVE into realization of P2P network,those problems could be resolved.

As oneof essentialcharacteristics of CVE,the usernormally

more interacts to hisneighbor 3D contents as manipilated andretrieved objects than the distant 3D contents. Therefore, byconnecting to the computers with 3D contents which are closeeach other in CVE, those 3D contents are efficiently obtained.When the 3D content retrieving is executed, queries are is-sued to the entire CVE space. In this case, by issuing queriesby using biased random walk [2], efficient information collec-tion can be attained although not so efficient as client/servermodel.Also such queryrequests are normally carried out whentheuserjoins tothe CVE, but notfrequentlybe issuedin wholesession.


550-445X/06 $20.00 © 2006 IEEE



In our system, by take advantage of those characteristics,P2P network can be configured using location information.Node is an element of P2P network and corresponds to a 3Dcontent in CVE with 1:1 relation. Since there are 3D contentsin CVE which are equivalent to the number of nodes, multiplenumbers of nodes are exited with a physical host computer.

A nodeis connectedto theneighbor nodewherethe distance

between nodes is mutually short. This connection is called alink. Therefore, P2P network is constructed by links betweennodes. Here, we assume there at most 4 links from a node inright directions as shown in Figure 3. For this reason, the linkconfiguration is dynamically changed as the 3D content movesin CVE. As query method to find the node with the 3D con-tent which is fit on the user’s query condition, biased randomwalk method is applied in our system. In summary, our Over-lay Network Layer has the following characteristics;

• 3D content corresponds to node by 1:1 relation,

• P2P network is organized by node location and link di-rection as shown in Figure 3,

• neighbor nodes with short distance arelinked each other,

• at most 4 links from a node in right directions are main-tained except the case where node does not exit in the

corresponding directions

• P2P network topology dynamically changes as 3D con-tents move.

Simulation result according to P2P network with the abovecharacteristics is shown in Figure 4. From this result we cansee how the P2P network is organized as a cobweb by the 100nodes which are distributed on the random locations.

Figure 3. Direction of Network

5. Contents Control Layer

Contents ControlLayermanages3D contents to be sent andreceived through Overlay Network Layer introduced in previ-

oussection.When new3D contents arerequested byuser, Con-tents Control Layer issues a request to Overlay Network andgains the target 3D contents. On the contrary, when the 3Dcontents managed by some user are requested by other user,thisrequestis alsoreceived throughOverlay Network andsent.The information that new 3D contents is added to CVE andmaintained on a node is transmitted to the neighbor nodesthroughOverlay Network.In orderto realizequickresponseforpresentation, those 3D contents are distributed to the neigh-bor nodes and managed individually. The 3D contents withthe required quality are obtained from Interest Control Layer

Figure 4. Simulation of P2P Network

depending on user’s interest. Here, visual quality of 3D con-tents is controlled by user’s interest level, called Level of Inter-est (LOD) as shown in Figure 5 and is precisely explained innext section.

Figure 5. Example of LOD(Level Of Detail) con-trol

6. Interest Control Layer

The user’s interest is analyzed by user’s operations to the3D contents and the relation between the 3D continents. Thus,by considering user’ interest, the interaction to the unneces-sary 3D contents can be avoided and eventually usability canbe improved. In this section, wedescribe modeling of user’sin-terest and reflecting to CVE system.

6.1. Model of user’s interest

• 3D Contents depicted by CVE3D Contents depicted by CVE is composed of the oneshown in figure 2. It is Avatar and Artifact becomingthe target of processing that uses interest information inthem. Avatar is 3D Contents that depicts the user on thesystem, and Artifact is 3D Contents that presentation isdone on the system.The matrix O of target 3D Contents for processing thatuses interest information is defined as follows. Avatar is


550-445X/06 $20.00 © 2006 IEEE



U andAvatar’s total is m. Artifact is P and Artifact’s to-tal is n.

O =

⎛⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎝

o1o2...

omo(m+1)o(m+2)

...o(m+n)

⎞⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎠

=

⎛⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎜⎝

U 1U 2

...U mP 1P 2

...P n

⎞⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎟⎠

Moreover, parameter C (curiosity) that sets the reflec-tiondegreeto interestinformationis definedin allAvatar.C isdecided by the user’sinput.As a result, the useris tobe able to adjust the system according to the purpose.

C =

⎛⎜⎜⎝

c1c2...

cm

⎞⎟⎟⎠

ci ∈ [0, 1]

At the following, “Contents” in sentences means bothAvatar and Artifact.

• Interest information matrixMatrix W is defined on the assumption of the definitionof the above-mentioned. W is the matrix of m× (m + n)(Total of Avatar × Total of Contents).

W =

⎛⎜⎜⎝

w11 · · · w1(m+n)

w21 · · · w2(m+n)

.... . .

...wm1 · · · wm(m+n)

⎞⎟⎟⎠

wij ∈ [0, 1]

wij shows the weight of the interest to Contents j fromAvatari. This system doesn’t consider the interest fromArtifact to Avatar.

• History matrix of interestWe aim at the achievement of the recommendationwith high accuracy by making the history of inter-est. A(ActiveHistory ) is history matrix of Contents inwhich Avatar have been interested. P (P assiveHistory)is history matrix of Avatar that have been inter-ested in Contents. A is the matrix of m × (m + n)(Total of Avatar × Total of Contents). aij is an or-der of the history of Contents j for Avatari. An initialvalue is 0.

A =

⎛⎜⎜⎝

a11 · · · a1(m+n)

a21 · · · a2(m+n)

.... . .

...am1 · · · am(m+n)

⎞⎟⎟⎠

aij ∈ [0, m + n]

P isthematrix of(m + n)×m (Total of Contents×Totalof Avatar). pij is an order of the history of Avatar j thatwas interested in Contentsi. An initial value is 0.

P =

⎛⎜⎜⎝

p11 · · · p1m p21 · · · p2m

.... . .

... p(m+n)m · · · p(m+n)m

⎞⎟⎟⎠

pij ∈ [0, m]

6.2. Reflection of interest information to

system

User’s interest information is updated in the followingcases.It reflects it in the systemby using theabove-mentionedinterest model.

• Interest acquisition from user’s action

User’s interest is acquired from user’s action in CVE. Aconcrete method classifies the action assumed with thesystem, and beforehand, defines weight in each action.

W

=

⎧⎪⎪⎪⎨⎪⎪⎪⎩

w

1 Action1

w

2 Action2

......

w

n Actionn

w

n ∈ [0, 1]

The user who does the action is assumed to be i. The tar-get contents of the interest of Useri are assumed to beContents j. The done action is assumed to be Actionn. Inthis case, wij is obtained by the following expressions. ciis degreeof theinterest acquisition by theaction that the

user sets.wij ← wij + w

n · ci

When it becomes wij ≥ 1 by useri’s action,ActiveHistory to useri’s Contents j becomes thefirst, andthe valueof aij becomes1. And, theorder of theentire ActiveHistory of the useri is updated. Addition-ally PassiveHistory to the useri of Contents j becomesthe first, and the value of pji becomes 1. And, the or-der of the entire P assiveHistory of the Contents j isupdated.

• RecommendationWhen the user was interested in contents (When theuser does the action), this system does the recommen-dation of Contents that seems suitable for the user byusing history of the interest. The history of the inter-est has ActiveHistory and PassiveHistory as shown

by the above-mentioned interest model, each element of ActiveHistory is composed of Contents, and each ele-ment of PassiveHistory is composed of Avatar.The user is assumed to be i. Avatar interesting of useri isassumed to be j. ActiveHistory to Contentsk of Avatar jis assumed to be ajk . At this time, the recommendationby ActiveHistory becomes the following expressions. ciis degree of the reflection of the interest by recommen-dation that the user sets. if the ActiveHistory element’svalue is 0 then this process is not done.

wik ← wik + wjk · ciajk (f or 1 ≤ k ≤ m + n)

The user is assumed to be i. Contents interesting of user i is assumed to be j. P assiveHistory to Avatark

of Contents j is assumed to be pjk At this time, the

recommendation by P assiveHistory becomes the fol-lowing expressions. ci is degree of the reflection of theinterest by recommendation that the user sets. if theP assiveHistory element’s value is 0 then this process isnot done.

wik ← wik + ck · cipjk (f or 1 ≤ k ≤ m)

Like this, the recommendation does Contents that be-came user’s interest object and relating Contents. Thispromotionsthat discoverunknown 3D Contents, and dis-cover new interest.


550-445X/06 $20.00 © 2006 IEEE



• Attenuation of interest by passage of timeIt is thought that the interest attenuates with the pas-sage of time.The user is renewed and weight procession W of T inter-val user’s interest is renewed in the place where the totalof Contents that exists in i and 3D virtual space is as-sumed tobe m + n. ci is a degreeof theattenuationof the

interestby thepassage of the time that user i sets. Atten-uation is expressed as follows.

wik ← wik − wik · ci (f or 1 ≤ k ≤ m + n)

• LOD(Level Of Detail) controlTheexpressionof Contentsby 3D virtual spaceis changedby W . The expression when useri sees Contents j in 3Dvirtual space. It changes according to the value of wij.This is called LOD(Level Of Detail) control. The expres-sion level is changed by the threshold in wij. It is possi-ble to express it as follows by the setting of the thresholdas t(threshold).

⎧⎪⎪⎨⎪⎪⎩

t0 ≤ W ij < t1 Level1t1 ≤ W ij < t2 Level2

.

.....

tn−1 ≤ W ij < tn Leveln

t0 < t1 < · · · < tn

Contents corresponding to the value of weight proces-sion W of user’s interest is acquired by Contents Con-trol Layer, and it displays it with Intaface Layer. As a re-sult, the expression(e.g figure5) of Contents according touser’s interestbecomes possible. Moreover, adjusting thelevel by arbitrarily setting the threshold becomes possi-ble.

7. Prototype System

We are developing a virtual shopping mall as one of appli-

cations using ourproposedCVE systemwhere two typesusersare existed, buyer and seller. The buyer exists as an avatarby walking through in shopping mall and looking for artifactsfrom shops in CVE. Theseller canoffer products forbuyers byallocating products on the self in his shop. Through P2P net-work organized by Overlay Network, scaleable virtual shop-ping mall with a number of users can b e realized.

Alsoby reflectingthe buyer’sinterest for shopping byInter-est Control Layer, the artifact with higher media quality withhigher user’s interestwhile withlow quality for lower user’s in-terest using LOD function, improvement of usability in CVEcan be expected.

8. Conclusion

Inthispaper,weproposedanetworkinfrastructureforCVE

that realizes scalable 3D VR space using P2P network, andmodeling and analyzing method of user’s interest. In our pro-posed P2P network, the 3D contents which are existed neara user can be regarded as his operational and retrieval objec-tives,whilethe3DcontentsfarfromhimcanbeneglectedintheCVE system. The nodes organized by P2P network can be re-garded 3D contents with 1:1 relation and can be distributedover manyphysical computersto realize scalable CVE system.By modeling and analyzing method user’s interest, the qual-ity of 3D contents can be controlled based on LOD function,eventually usability can be improved.

Now we are developing a virtual shopping mall as one of our proposed CVE system to evaluate the usefulness. As fu-tureworks,user’s interestanalysismethod willbe extendedbyaddingpreciseprofile information with users such as ages, sex-ual and profession information and attributes information of artifacts, such as statistical information of consumers on realmarket.

References

[1] Y.ArakiandS.M. E.J. Inc. Vsplus:A high-levelmulti-userextension library for interactive vrml worlds. In VRML

’98: Proceedings of the third symposium on Virtual reality

modeling language, pages 39–47. ACM Press, 1998.

[2] Y. Chawathe, S. Ratnasamy, L. Breslau, N. Lanham, andS. Shenker. Making gnutella-like p2p systems scalable.In SIGCOMM ’03: Proceedings of the 2003 conference on

Applications, technologies, architectures, and protocols for

computer communications, pages 407–418. ACM Press,2003.

[3] J. X. Chen, Y. Yang, and B. Loftin. Muvees: a pc-basedmulti-user virtual environment for learning. In VR ’03:

Proceedings of the IEEE Virtual Reality , pages 163–170,

2003.[4] D.Ding andM. Zhu. A model of dynamic interestmanage-

ment: interaction analysis in collaborative virtual environ-ment. In VRST ’03: Proceedings of the ACM symposium

on Virtual reality software and technology , pages 223–230.ACM Press, 2003.

[5] C. Greenhalgh, J. Purbrick, and D. Snowdon. Insidemassive-3: flexible support for data consistency and worldstructuring. In CVE ’00: Proceedings of the third interna-

tional conference on Collaborative virtual environments,pages 119–127. ACM Press, 2000.

[6] J. Kamahara, T. Asakawa, S. Shimojo, and H. Miyahara.A community-based recommendationsystem to reveal un-expected interests. In MMM ’05: Proceedings of 11th In-

ternational Multimedia Modelling Conference, pages 433–

438, 2005.[7] H. Nakanishi, S. Koizumi, T. Ishida, andH. Ito. Transcen-dent communication: location-based guidance for large-scale public spaces. In CHI ’04: Proceedings of the 2004

conference on Human factors in computing systems, pages655–662. ACM Press, 2004.

[8] S. Oikawa. Qos control function based on user’s informa-tionon the3d virtualsharedspace. In AINA’04:18th Inter-

national Conference on Advanced Information Networking

and Applications, page 74, 2004.


550-445X/06 $20.00 © 2006 IEEE

Documents

Scalable Collaborative Virtual EnvironmentConsidering User’s Interest Information