Web 2.0 Services for Identifying Communities of Practice through ...

Web 2.0 Services for Identifying Communities of Practice through Social Networks Stephen J.H. Yang*, Jia Zhang**, Irene Y.L. Chen*** *National Central University, Taiwan **Northern Illinois University, USA ***Ching Yun University, Taiwan *[email protected], **[email protected], ***[email protected] Abstract This paper presents a social network-based peer-topeer search service for identifying right collaborators in the context of Web 2.0. We present a three-layer hierarchical social network, in which we identify two important relationship ties –knowledge relationship tie and social relationship tie. These relationship ties are metric used to measure the collaboration strength between pairs of participants on a social network. The stronger the knowledge relationship tie, the more knowledgeable the participants; the stronger the social relationship tie, the more likely the participants are willing to share their knowledge. By analyzing and calculating these relationship ties among peers using our computational model, we propose a systematic way to discover collaboration peers according to configurable and customizable requirements. Experiences of providing Web 2.0 services for identifying communities of practice through peer-to-peer search are also reported. Keywords: Web 2.0, social networks, communities of practice, peer-to-peer

1. Introduction Web 2.0 represents a collection of the next-generation technologies over the Internet. Its most significant features center around “ c ompl e t eopenn e s s ”in four aspects [1]. First, Web 2.0 introducesac on c e ptof“ ope nc ommu n i t y , ” where any Internet users can participant in Web 2.0 to collaborate in a loosely regulated manner for a common goal. Second, any collaboration work goes through online approaches and becomes available to all participants instantaneously. Third, Web 2.0 is mainly built on open standards. Fourth, Web 2.0 encourages leverage of opensource software. In short, Web 2.0 provides an unprecedented platform for all Internet users to dynamically form collaboration groups and create, publish, exchange, share, and cooperate on any types of information (knowledge). Apparently, the success of Web 2.0 heavily relies on communication and collaboration among people over the Internet - what people possess; whether people are willing

2007 IEEE International Conference on Services Computing (SCC 2007) 0-7695-2925-9/07 $25.00 © 2007

to communicate; how a group of people can be formed as communities of practice; and so on. Meanwhile, Web 2.0 enables and facilitates Internet collaboration by providing a cluster of new generation of social software such as Wikis [2], Blogs [3], Really Simple Syndication (RSS) feeds [4], video podcast, Ajax [5]-based browsers, peerto-peer, instant messenger, and other social networking software. Some successful examples of Web 2.0 applications are Wikipedia [6], YouTube [7], MySpace [8], and Flickr [9]. In short, Web 2.0 is widely considered as the next generation of Web technologies and applications [10]. One of the essential goals of applying Web 2.0 services in communities of practice is to enhance interactive communication and collaboration among participants in the communities. By communities of practice, we refer to a group of participants with common interests in a particular subject [11]. By participants, we refer to the individuals who (1) possess related information, (2) can help to discover and obtain the information, or (3) are willing to exchange and share information with others. The performance of collaboration is fundamentally based on how proper participants can be effectively found. As a result, one of the critical challenges of Web 2.0 services is how to identify the right participants to form communities of practice. It should be noted that in Web 2.0, participants are readers as well as writers: they are both consumers and producers of information (knowledge) and services. This paper presents a three-layer social network-based peer-to-peer (P2P) search service to facilitate identification and establishment of communities of practice in the context of Web 2.0. Here peers represent individuals (participants) who are associated with the communities by knowledge and social relationships. Th r ou g h ou tt h i spa pe r ,wewi l lu s et h et e r ms“ pe e r ”a n d “ p a r t i c i pa n t s ”i n t e r c h a n g e a bl y .We propose two important relationship ties, knowledge relationship tie and social relationship tie, as underlying metric to measure the degrees of pe e r s ’knowledge matching with a query, as well as the degrees of social relationships among peers. By analyzing and calculating these relationships among peers using our computational models, we present a

systematic way to discover peers based on configurable and customizable requirements. Furthermore, we have conducted experiments to evaluate how to improve the identification of communities of practice in Web 2.0 with such a social network-based P2P framework. The remainder of the paper is organized as follows. In Section 2, we review related work. In Section 3, we introduce our social network-based P2P framework. We present the methods for calculating knowledge relationship tie and social relationship tie in Section 4 and Section 5, respectively. In Section 6, we discuss our experiments and results. In Section 7, we draw conclusions.

2. Related work Social networks are built upon a hypothesis that there exists a determinable networking structure of how people know each other [11]. In such networks, people are connected through common social relationships, either directly or indirectly [12]. Researchers have recognized that a broader sense of social network is a self-organized structure of people, information, and communities [13, 14]. A social network thus can be formalized into a net structure comprising nodes and edges. Nodes represent individuals or organizations. Edges connecting nodes are called ties, which represent the relationships between the individuals and organizations. The strength of a tie indicates how strong the relationship is. Many kinds of ties may exist between nodes [12]. Social networks can also be represented as matrices, so that their properties can be analyzed using graph theory. In this paper, we will address knowledge relationship tie and social relationship tie. Considering the maximum size of a social network tends to contain around 150 nodes [15], in our research, we consider the scope of a social network with about 50 to 150 peers within a university. Social interaction ties are the structural links created through the social interactions between individuals in a network [16-19]. Prior studies sugges tt h a ta ni n di v i du a l ’ s centrality in an electronic network of practice can be measured using the number of social ties that an individual has with others in the network [20]. Some researchers have addressed the importance of social interaction ties in knowledge exchange. For example, Tsai and Ghoshal [21] report that social interaction tie has positive impacts on the extent of inter-unit resource exchange. Wasko and Faraj [18] discover that the centrality established by the social interaction ties significantly impacts the helpfulness and volume of knowledge contribution. A P2P network is a distributed networking structure that treats every participant as a peer and allows each peer to play as either a client or a server under different circumstances [22, 23]. P2P and social networks share

2007 IEEE International Conference on Services Computing (SCC 2007) 0-7695-2925-9/07 $25.00 © 2007

many concepts in common. For example, they are both distributed networking structures; a peer in a P2P network can be viewed as an analog of a node in a social network; a link in a P2P can be viewed as an analog of a relationship tie in a social network. In contrast to most P2P searches that emphasize on search queries and protocols, our social network-based P2P search aims at reducing search time and decreasing message traffic by minimizing the number of messages circulating in the network.

3. Social network-based P2P framework The key idea of our social network-based P2P framework is illustrated in Figure 1. For a given query requesting participants with certain knowledge, a social network containing relevant participants with the requested knowledge will be dynamically constructed within the scope of a P2P network.

Figure 1. A three-layer social network-based P2P Our framework comprises three dynamically created layers. We will use an example to illustrate what the layers are and how to construct them. As shown in Figure 1, the first layer is the P2P knowledge net (K-net), which is established to connect available peers who own the requested knowledge into a pool of active peers. As mentioned, we confine the scope of the pool to a smallscale P2P network with about 50 to 150 peers within a university. A peer in the pool can be either a knowledge repository or a knowledgeable individual. A weighted edge between two peers is called a knowledge relationship tie, which is used to measure the degree of how a pe e r ’ s knowledge matches the other’ squ e r y . For example, if a peer (e.g., peer Steve) makes a request for peers with “ Software Engineering”k n owl e d g e , a P2P K-net will be dynamically generated based on the query. As shown in Figure 1, peers Chris and Albert are with knowledge relationship ties (0.8) and (0.16), respectively, which means Chris’k n owl e dg e ma t c h e sb e t t e rwi t h Steve’ s

Cognitive Process Dimensions Knowledge dimension

Remember

Understand

Apply

Analyze

Evaluate

Create

Factual knowledge

0.9

0.8

0.4

0.4

0

0

Conceptual knowledge

0.3

0.3

0.3

0.1

0

0

Procedural knowledge

0.6

0.5

0.3

0.2

0

0

Metacognitive knowledge

0

0

0

0

0

0

Figure 2(a). Example of Albert’ sBloom taxonomy matrix request than Albert’ s . P2P social net (S-net) is the second layer. A weighted edge between two peers in S-net represents a social relationship tie, which is used to measure the degree of social familiarity between the two peers. Peers on K-net without the requested knowledge will be removed from Snet (e.g., Mary). Using the example shown in Figure 1, Steve is more familiar with Albert than Chris because the social relationship tie between Steve and Albert is (0.9), which is greater than that between Steve and Chris (0.8). Based on the generated S-net, an instant messengerequipped group discussion, shown as the third layer in Figure 1, is invoked to help Steve communicate with the peers found in S-net (Chris and Albert). Peers appear on S-net with negative relationship with the requester will be removed from the group discussion (e.g., Bob). This example shows that the essential challenge of constructing this three-layer social network is how to calculate knowledge relationship tie and social relationship tie.

domain. As shown in Figure 2(a) and 2(b), Bloom taxonomy is a matrix consisting of two dimensions: Knowledge dimension and Cognitive Process dimension. The former indicates the types of knowledge; the latter indicates cognitive processing of knowledge. Each cell in the matrix is associated with a value ranging from 0 to 1, representing the level of proficiency. For example, let Figure 2(a) and Figure 2(b) indicate peer Albert and peer Christ’ sk n owl e dg epr of i c i e n c yr e g a r d i n gt h ek n owl e dge doma i n of “ Software Engineering,” r e s pe c t i v e l y .I n Figure 2(a), the cell (Factual knowledge, Remember) has a value (0.9), which indicates that Albert is good at me mor i z i n g f a c t u a l k n owl e dg e a bou t “ Software Engineering.” I n Fi g u r e 2( b) ,t h ec e l l( Con c e ptual knowledge, Apply) has a value (1.0), which indicates Chris is excellent at applying conceptual knowledge of “ Software Engineering.” Consider peer i is requesting peer j whose knowledge proficiency conforms to a requested knowledge domain k. Peer i’ squery can be calculated by:

4. Calculation of knowledge relationship tie Without losing generality, in our research, we consider ape e r ’ sk n owl e d g edoma i n ,pr of i c i e n c y ,a n dr e pu t a t i onof contribution as key indicators determining its capability to participate in collaborations. Therefore, as shown in K-net in Figure 1,wec a l c u l a t eap e e r ’ sk n owl e dg er e l a t i on s h i p tie based on the three indicators.

y e KQ( k ) (i, j ) K (proficienc ( j ) K (conformanc (i )T k) k)

KQ(serialized (i, j ) = k)

6   KQ( k ) (m, n)   m 1 n 1  4

K (tiek) (i, j ) KQ (serialized (i, j ) K (reputation ( j) k) k) KQ( k ) (i, j ) is a Bloom taxonomy matrix representing

4.1. Rationale and equations

a query by peer i, which is requesting for peer j whose knowledge proficiency conforms to a requested

We use Bloom taxonomy matrix [24] to classify a p e e r ’ sdoma i nk n owl e dg ea n di t sproficiency in such a

Cognitive Process Dimensions Knowledge dimension

Remember

Understand

Apply

Analyze

Evaluate

Create

Factual knowledge

0.2

0.7

0.7

0.7

0.8

0.8

Conceptual knowledge

0.8

0.7

1.0

0.8

0.7

0.7

Procedural knowledge

0.2

0.2

0.3

0.2

0

0

0

0

0

0

0

0

Metacognitive knowledge

Figure 2(b). Example of Chris’ sBloom taxonomy matrix

2007 IEEE International Conference on Services Computing (SCC 2007) 0-7695-2925-9/07 $25.00 © 2007

knowledge domain k.

K

proficiency (k )

( j)

y K (proficienc ( Albert) be the matrix shown in Figure 2(a), SE )

is

a

Bloom

taxonomy

matrix

representing peer j’ sk n o wl e dg e proficiency w.r.t. a requested knowledge domain k. e K (conformanc (i) k)

is a Bloom taxonomy matrix

representing a conformance requirement requested by peer i to peers whose knowledge proficiency conforms to a requested knowledge domain k.

KQ(serialized (i, j ) is the serialization of KQ( k ) (i, j ) , k) which is a 6 by 4 Bloom taxonomy matrix.

K (tiek) (i, j )

is a real number between 0 and 1

representing the knowledge relationship tie between peer i and peer j w.r.t. knowledge domain k.

then:

KQ( SE ) ( Steve, Albert) y e K (proficienc ( Albert ) K (conformanc ( Steve)T SE ) SE )

0.9 0.8 0.4 0.4   0.3 0.3 0.2 0.1 =   0.6 0.5 0.3 0.2  0 0 0 0 0 0 

0

0 0 0

 0 0  0 0 

0 0

0 0 0  0 0 0

After serialization,

KQ(serialized (Steve, Albert) = k)

4.2 Examples and discussions The value of

K (tiek) (i, j )

indicates the degree of peer

j’ sk n o wl e dg ema t c h e speer i’ squ e r y :t he higher the value, the stronger the tie. For example, consider a peer, Steve, who requests for peers having k n owl e dg e“ Software Engineering”wi t ht h epr of i c i e n c yofa pp l y i n gc on c e pt u a l k n owl e dg eof“ Software Engineering.”S t e v e ’ srequest can be denoted as K

conformance ( SE )

( Steve) . Based on the

aforementioned equations and the example shown in Figure 1 and 2, we found two peers, Albert and Chris, whose knowledge relationship ties,

Moreover, than

since

K

tie ( SE )

( Steve, Chris ) is greater

K (tieSE ) ( Steve, Albert )

,

Chris

is

more

knowledgeable than Albert in terms of meeting Steve’ s r e qu e s tofa ppl y i n gc o n c e pt u a lk n owl e d g eof“ Software Engineering.” Th e de t a i lc a l c u l a t i on of k n owl e dg e relationship tie is illustrated as follows: Let Steve’ sr e q u e s tbe :

0   0 e K (conformanc ( Steve)T  SE )  0  0  Let

Albert’ s

T

0 0 0 0 0 0 1 0 0 0  0 0 0 0 0  0 0 0 0 0

k n owl e dg e

pr of i c i e n c y ,

2007 IEEE International Conference on Services Computing (SCC 2007) 0-7695-2925-9/07 $25.00 © 2007

6   KQ( k ) (m, n) = 0.2  m 1 n 1  4

Let Albert’ sk n owl e dg er e pu t a t i onb e reputation K ( SE ) ( Albert ) = 0.8, then

K (tiek) ( Steve, Albert )

KQ(serialized ( Steve, Albert ) K (reputation ( Albert ) k) SE ) = 0.2 0.8 = 0.16 As a result, Steve’ sk n owl e dg er e l a t i o n s h i pt i et oAlbert is 0.16.

K (tieSE ) ( Steve, Albert ) and K (tieSE ) (Steve, Chris) are greater than zero, which means both of them conform to S t e v e ’ s request in terms of knowledge relationships.

T

0 0 0 0 0 0 1 0 0 0  0 0 0 0 0  0 0 0 0 0

=  0 0 0.2 0 0 0  

K (reputation ( j ) is a real number between 0 and k) 1representing peer j’ sr e pu t a t i onr e g a r di n gc on t r i bu t i ont o the requested knowledge domain k.

0 0  0  0 0 0  0 0  0  0 0  0

Let

K

Chris’ s

proficiency ( SE )

k n owl e dg e

pr of i c i e n c y ,

(Chris) be the matrix shown in Figure 2(b),

then the query of calculating the degree of Albert’ s proficiency conforms to Steve’ sr e qu e s ti sde n ot e da s :

KQ( SE ) ( Steve, Chris) y e K (proficienc (Chris ) K (conformanc ( Steve) T SE ) SE )

0.2  =  0.8   0.2  0 0 0  =  0 0   0 0  0 0 

0.7 0.7 0. 2 0 0 1 0 0

0.7 1 0. 3 0 0 0 0 0 0 0 0 0

0.7 0.8 0.2 0 0 0  0  0

After serialization,

0.8 0.8  0 0 0  0.7 0.7  0 0 1  0 0  0 0 0  0 0  0 0 0

0 0 0 0

0 0 0 0

T

0 0  0  0

4 6  KQ(serialized ( Steve, Chris) =   KQ( k ) (m, n) = 1.0 k) m 1 n 1 

Let Christ’ sknowledge reputation be

K (reputation (Chris ) = 0.6, then SE ) K(tiek) (Steve, Chris)

KQ(serialized (Steve, Chris) K(reputation (Chris) k) SE) = 1.0 0.6 = 0.6 As a result, Steve’ sk n owl e dg er e l a t i on s h i pt i et oChris is 0.6. This example illustrates that Chris is more knowledgeable than Albert in terms of helping Steve to a p pl yc on c e pt u a lk n owl e d g eof“ Software Engineering.”

5. Calculation of social relationship tie The social relationship tie indicates the degree of social familiarity between pairs of peers on the S-net. For a pair of peers, denoted by peer i and peer j, the social relationship tie between them is the product of their social familiarity, social reputation, and social trust. Huhns and Buell [25] indicate that people are more likely to trust something that has been proved. Similarly, people are more likely to collaborate with someone with good reputation. Many approaches have been proposed to calculate the degree of trust (reputation) based on experiences, such as rating mechanism and referral networks [26]. In our previous research, we have utilized t h er a t i n gme c h a n i s mt oe v a l u a t eWe bs e r v i c e s ’r e pu t a t i on . We apply the same rating mechanism in this research to e v a l u a t ep e e r s ’r e pu t a t i on . Due to length limitation, interested readers please refer to [26] for detailed discussions about our calculations using the rating mechanism regarding trust and reputation.

S tie (i , j ) S familiarit y (i, j ) S reputation ( j ) S trust (i, j ) where

S tie (i, j )

is a social relationship tie between peer i

and peer j,

S familiarity (i, j )

is a social familiarity between peer

i and peer j, ssocial reputation. S reputation ( j ) is peer j’

social network should specify its social familiarity with a new peer first connecting to the social network, by filling forms and answering questionnaires. If an existing peer does not specify its social familiarity with a new peer, the default value is zero meaning that there is no relationship between them. Social familiarity can exhibit different levels of familiarity relationships, such as friends, teammates, organization colleagues, or virtual community members. Meanwhile, social familiarity can be either positive or negative values ranging between -1 and 1, indicating the relationship is either good or bad. To perform quantitative analysis, without losing generality, we define social familiarity between peers i and j into the following nine categories:

S familiarit y (i, j ) = 0, if there is no relationship between peer i and peer j;

S familiarit y (i, j ) = 0.8~1.0, if peer i considers peer j a friend with positive relationship; S familiarit y (i , j ) = 0.5~0.7, if peer i considers peer j a team-mate with positive relationship;

S familiarit y (i, j ) = 0.3~0.4, if peer i considers peer j an organization colleague with positive relationship;

S familiarit y (i, j ) = 0~0.2, if peer i considers peer j a virtual community member with positive relationship;

S familiarit y (i , j ) = -0.8~-1.0, if peer i considers peer j a friend with negative relationship;

S familiarit y (i, j ) = -0.5~-0.7, if peer i considers peer j a team-mate with negative relationship;

S familiarit y (i, j ) = -0.3~-0.4, if peer i considers peer j an organization colleague with negative relationship;

S familiarit y (i , j ) = 0~-0.2, if peer i considers peer j a virtual community member with negative relationship

5.2. Social reputation Each peer has a social reputation, which is the product oft h epe e r ’ ss oc i a lr a t i n g[ 10]a n dt h ea v e r a g eoft h e pe e r ’ ss oc i a lf a mi l i a r i t i e s .So c i a lr e p u t a t i o nr e pr e s e n t sa degree of confidence to a target peer from all other peers on a social network who know the target peer. The social reputation of peer j is computed as follows:



= Social familiarity indicates the level of familiarity ranging from casual to close. Each peer existing in a

2007 IEEE International Conference on Services Computing (SCC 2007) 0-7695-2925-9/07 $25.00 © 2007

S

NoP ( m )

S trust (i, j ) is the degree of trust of peer i to peer j 5.1 Social familiarity



S reputation ( j ) = AVG S familiarity ( j ) S rating ( j )

where:



familiarity

( j, m)

m 1

NoP( j )





S rating ( j ) ,

AVG S familiarity ( j ) is an average value of

peer j’ ss oc i a lf a mi l i a r i t i e s ,

S

rating

( j ) is peer j’ ss oc i a lr a t i n g ,

NoP(j) is the number of peers connected to peer j.

5.3. Social trust Social trust is a confidence of how a pair of peers on the social network treats each other. It also indicates how a peer is associated with other peers directly connected to it on the social network. For a pair of peers who are socially related, as denoted by the requesting peer i and the requested peer j, the trust association between them is denoted by

S trust (i, j ) ,

indicating the confidence of

trustworthiness from peer i to peer j. S trust (i, j ) is used to determine whether the requested peer conforms to the r e qu e s t i n gp e e r ’ srequirements of trustworthy. The value trust

of S (i, j ) is denoted by percentage: the higher the confidence is, the higher the trust association is. For example, if the value of S trust (Chris , Albert ) is 78%, it means that the requesting peer Chris has 78% confidence that the requested peer Albert is trustworthy. We utilize sampling of binomial probability to calculate the value of S trust (i , j ) , based on a 95% confidence interval in terms of probability [27]. First, we define the following terms: S is a set of interaction instances representing samples of t h e r e qu e s t e d pe e r ’ s past interactions, . S  s1 , s 2 ,.... s n  Tr is a set of trust evaluation values containing past experience instances. It is denoted by . Tr  tr1 , tr2 ,....trn 

Rating : S  Tr Rating s : The Rating function maps the interaction instance s to past experience

instance, tr . In other words, the function associates past service instances with past experience instances. Experiences are collected by requested peers rather than requesting peers. Accpet : Tr   0,1A requirement hypothesis can be denoted as an Accept function. The output of an Accept function is 1 when past experience instance is accepted by the requesting peer, and 0 otherwise. 1 Accept  tr  0

Accept otherwise

Based on the usage of Large-Sample of Hypothesis for a Binomial Proportion to evaluate the simple error and true error of a hypothesis addressed in [27, 28], the result of the hypothesis assesses the sample is a Boolean value (true or false). Thus, we can see that the hypothesis assesses the sample as a Bernoulli trial, which distribution

2007 IEEE International Conference on Services Computing (SCC 2007) 0-7695-2925-9/07 $25.00 © 2007

is a binomial distribution. A binomial distribution approximates a normal distribution when the number of sample is large enough. Simple error represents correct rate in samples and true error represents correct rate in population. We can then obtain a confidence interval according to the simple error. The area of confidence interval represents a probability of true error falling in the interval. In a normal distribution, the true error is 95% probabilities falling within the range of mean 1.96 SD (Standard Deviation) in compliance with the experience rule. In other words, we can utilize the confidence interval to evaluate the lowest true error of the evaluating hypotheses. Let Accept function be the hypothesis; we can evaluate the possible true error of the hypothesis based on the past instances S according to the Evaluating Hypotheses theory (Mitchell, 1997). Whether the tr ( tr E ) is accepted by Accept is a binomial distribution, which approximates the normal distribution when the number of samples is large enough. Thus, we can utilize the normal distribution to calculate that the sample error closes with the true error. The true error is of 95% probabilities falling within a confidence interval, which will be approved as a trustworthy peer in the general application. We define the confidence symbol as the lowest bound of the true error. The trust of service conforms to the r e qu e s t ’ sr e qu i r e me n twh e nt h ec on f i de n c ei sh i g h e r . pˆ =

1 Accept Rating ( s) , n s ∈S

SD 

, pˆ 1 pˆ n

z 95% 1.96

Confidence max pˆz 95% SD, 0 As the number of samples increases, the standard deviation decreases and the confidence will become closer to the true error. For example, assume the past instances of a requested peer is denoted as S, and let S 256 . The requesting leaner proposes a Requirement Hypothesis Accept. If the result of calculation is pˆ0.6 , the confidence can be calculated from the following equation: 1 , pˆ Accept Rating(s) 0.6 256 sS

z 95% 1.96 Confidence pˆz 95% 

 pˆ 1 pˆ 256

0.6 0.060012 0.539987 trust

The calculated confidence S (i, j ) is 53.99%, which means the requesting peer has 53.99% confidence

that the requested peer can meet the trustworthy requirement based on 95% confidence interval. Hence, we can assert that the trustworthiness of the requested peer is 56.83% (53.99% over 95%) conforming to the requesting p e e r ’ srequirements.

5.4 Examples and discussions Based on the aforementioned equations and the examples shown in Figure 1, the detailed calculation of social relationship tie is illustrated as follows: Let: Albert

be

a

good

friend

of

Steve,

S familiarit y ( Steve , Albert ) 0.9 ; Albert’ ss oc i a lr a t i n gbe0. 6 ; Albert has three peers directly connected to him (Steve, Chris, and Bob); Steve’ sde g r e eoft r u s tt oAlbert is 0.57. Then Albert’ ss oc i a lr e p u t a t i o ni s :

S

reputation

S

NoP ( 3)

=

(Albert )

familiarity

( Albert,3)



S rating ( Albert )

m 1

NoP( Albert )

=

S

familiarity ( Albert, Steve) S familiarity ( Albert , Chris ) S familiarity ( Albert , Bob)

3

S

rating ( Albert )

= 0.9 0.8 (0.2) 0.6 = 0.3 3 So Steve’ ss oc i a lr e l a t i on s h i pt i et oAlbert is: tie

S ( Steve, Albert ) S

familiarit y

( Steve , Albert ) S

reputation

( Albert ) S

trust

( Steve , Albert )

0.9 0.3 0.57 0.153

Let Chris be a good friend of Steve,

S ( Steve, Chris ) 0.8 ; Christ’ ss o c i a lr a t i n gbe0. 8 ; Chris also has three peers directly connected to him (Steve, Albert, and Bob); Steve’ sde g r e eoft r u s tt oChris is 0.57. Then Chris’ ss oc i a lr e pu t a t i oni s : familiarity

S

reputation

S

NoP( 3)

=

(Chris )

familiarity

(Chris,3)



S rating (Chris)

m 1

NoP(Chris)

= S familiarity (Chris, Steve)S familiarity (Chris, Albert)S familiarity (Chris, Bob)Srating(Chris) 3

2007 IEEE International Conference on Services Computing (SCC 2007) 0-7695-2925-9/07 $25.00 © 2007

= 0.8 0.8 0.4 0.8 = 0.52 3

So Steve’ ss oc i a lr e l a t i on s h i pt i et oChris is tie

S ( Steve, Chris) S familiarity ( Steve, Chris) S reputation (Chris) S trust ( Steve, Chris) 0.8 0.52 0.57 0.237

This example concludes that Steve has stronger social relationship tie to Chris than to Albert. We can also infer that Chris is more likely to share his knowledge with Steve.

6. Experiments and discussions For quantitative performance evaluation, we measured two indexes: Precision and Recall. Precision is the fraction of the found peers who are considered as relevant; Recall is the fraction of the relevant peers who have been found. Precision and Recall are formally defined as follows: Precision = Ra , Recall = Ra A R Where: A contains a set of peers found, |A| is the number of peers in A. R contains a set peers found that are considered relevant, |R| is the number of peers in R. Ra contains a set of peers as the intersection of the sets R and A. |Ra| is the number of peers in Ra. 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

K-Tie (Precision) S-Tie (Precision) K-Tie (Recall) S-Tie (Recall)

Internet computing

Web computing

Mobile Internet

Wireless Web

Figure 3. Precision and Recall as performed by knowledge relationship tie (K-Tie) and social relationship tie (S-Tie) In this experiment, we used four kinds of domain knowledge as the search domains: Internet computing, Web computing, Mobile Internet, and Wireless Web. As indicated in Figure 3, for the four given search domains, the Precision of social relationship tie (S-Tie) search outperforms the knowledge relationship tie (K-Tie). This indicates that the found peers are more relevant and they are more likely to be in the same social group because they have higher social relationship ties. In contrast, the

Recall of K-Tie search outperforms S-Tie. This indicates that peers with the same knowledge domain will most likely to be found at the same time because they have higher knowledge relationship tie.

7. Conclusions The major contribution of this paper is applying social network technique to improve P2P search by finding knowledgeable and socially related participants in Web 2.0. In this paper, we have presented a three-layer social network-based P2P network equipped with the calculation methods of knowledge relationship tie and social relationship tie. Through such a social network-based P2P network, we demonstrated a new possibility of using social network to enhance P2P so that query can be routed to [29] peers with stronger relationship ties. Our research results can be utilized to implement a utility service in Web 2.0 platform to help dynamically establish collaboration groups effectively and efficiently based on particular topics. We see several areas that deserve further research. Peers and other participants may have their own needs when they search for participants and interact with others; therefore, we need to conduct further study on new relationship ties and investigate special requirements from different social perspectives in addition to knowledge and social relationships. It is also necessary to take into account collaboration context during the identification of communities of practice in Web 2.0.

Acknowledgements This work is supported by National Science Council, Taiwan under grant NSC95-2520-S-008-006-MY3.

References [1] L.-J. Zhang, J. Zhang, and H. Cai, Services Computing. 2007: Springer. [2] Wikis, Available from: http://www.wikipedia.org/. [3] Blog, Available from: http://en.wikipedia.org/wiki/Blog. [4] M. Pilgrim, "What is RSS?" 2002, Available from: http://www.xml.com/pub/a/2002/12/18/dive-into-xml.html. [5] Sun, Available from: http://developers.sun.com/ajax/index.jsp?cid=59754. [6] Wikipedia, "Wikipedia," Available from: http://www.wikipedia.org/. [7] YouTube, "YouTube," Available from: http://www.youtube.com/. [8] MySpace, "MySpace," Available from: http://www.myspace.com/. [9] Flickr, "Flickr," Available from: http://www.flickr.com/. [10] T. O'Reilly, "What Is Web 2.0, Design Patterns and Business Models for the Next Generation of Software," 2005, Available from: http://www.oreillynet.com/pub/a/oreilly/tim/news/2005/09/30/w hat-is-web-20.html. [11] H. Alani, S. Dasmahapatra, K. O'Hara, and N. Shadbolt,

2007 IEEE International Conference on Services Computing (SCC 2007) 0-7695-2925-9/07 $25.00 © 2007

"Identifying Communities of Practice through Ontology Network Analysis," IEEE Intelligent Systems, Mar.-Apr., 2003, 18(2): pp. 18-25. [12] E.F. Churchill and C.A. Halverson, "Social Networks and Social Networking," IEEE Intelligent Systems, Sep.-Oct., 2005, 20(5): pp. 14-19. [13] P. Raghavan, "Social Networks: From the Web to the Enterprise," IEEE Internet Computing, Jan./Feb., 2002, 6(1): pp. 91-94. [14] H. Kautz, B. Selman, and M. Shah, "ReferralWeb: Combining Social Networks and Collaborative Filtering," Comm. of ACM, Mar., 1997, 40(3): pp. 27-36. [15] R. Hill and R. Dunbar, "Social Network Size in Humans," Human Nature, 2002, 14(1): pp. 53-72. [16] R.S. Burt, Structural Holes: The Social Structure of Competition. 1992: Harvard University Press, Cambridge, MA. [17] R. Putnam, Tuning in, Tuning out: The Strange Disappearance of Social Capital in America, in Political Science and Politics. 1995. p. 664-683. [18] M.M. Wasko and S. Faraj, "Why Should I Share? Examining Social Capital And Knowledge Contribution in Electronic Networks Of Practice," MIS Quarterly, 2005, 29(1): pp. 35-57. [19] G. Zhang, Q. Jin, and M. Lin, "A Framework Of Social Interaction Support for Ubiquitous Learning," in Proceedings of the 19th International Conference on Advanced Information Ne t wor k i ngandAppl i c at i ons( AI NA’ 05) , 2005. [20] M. Ahuja, D. Galletta, and K. Carley, "Individual Centrality And Performance in Virtual R&D Groups: An Empirical Study," Management Science, 2003, 49(1): pp. 21-38. [21] W. Tsai and S. Ghoshal, "Social Capital And Value Creation: The Role Of Intrafirm Networks," Academy of Management Journal, 1998, 41(4): pp. 464-476. [22] J. Brase and M. Painter, "Inferring Metadata for a Semantic Web Peer-to-Peer Environment," Educational Technology & Society, 2004, 7(2): pp. 61-67. [23] K. Aberer, M. Punceva, M. Hauswirth, and R. Schmidt, "Improving Data Access in P2P Systems," IEEE Internet Computing, 2002, 6(1): pp. 58-67. [24] L.W. Anderson, D.R. Krathwohl, P.W. Airasian, K.A. Cruikshank, R.E. Mayer, P.R. Pintrich, J. Raths, and M.C. Wittrock, A Taxonomy for Learning, Teaching, and Assessing: ARe v i s i onofBl oom’ sTax onomyofEduc at i onalObj e c t i v e s . 2001: New York, Longman. [25] M.N. Huhns and D.A. Buell, "Trusted Autonomy," IEEE Internet Computing, May, 2002, 6(3): pp. 92-95. [26] S.J.H. Yang, J.S.F. Hsieh, B.C.W. Lan, and J.Y. Chung, "Composition and Evaluation of Trustworthy Web Services," International Journal of Web and Grid Services, 2006, 2(1): pp. 5-24. [27] T. Mitchell, Machine Learning. 1997, WCB McGraw-Hill. p. 128-141. [28] W. Mendenhall and R.J. Beaver, Introduction to Probability and Statistics. 1999, Duxbury Press. p. 442-446. [29] A. Andrieux, K. Czajkowski, A. Dan, K. Keahey, H. Ludwig, T. Nakata, J. Pruyne, J. Rofrano, S. Tuecke, and M. Xu, "Web Services Agreement Specification (WS-Agreement)," 2005, Sep. 20, 2005, Available from: http://www.ggf.org/Public_Comment_Docs/Documents/Oct2005/WS-AgreementSpecificationDraft050920.pdf.