Citation Context

...8-1-4244-9921-2/11/$26.00 ©2011 IEEE 876 capture structural equivalence, i.e., same profile of relations to all other nodes in the network, such as the Pearson correlation and the Jaccard coefficient =-=[5]-=-. Related work: In the literature, algorithms for detecting community structure have largely been applied to a given network structure, usually modeled as a graph. The most prominent algorithm thereof...

Citation Context

...ysis and assessment of these options. For the derivation of graph edge weights, we consider two metrics: the Proportional Similarity [20] and the inverse of the symmetrized KullbackLeibler divergence =-=[12]-=-. Effectively, each metric could be seen as a different transformation from one data set (distribution of user interests over interest classes) to another (graph edge weights). Comparing the outcomes ...

Citation Context

... internally by high-weight edges and to other subsets’ nodes with smallor zero-weight edges. These algorithms also assess, in the same time, the quality of this grouping through the modularity metric =-=[16]-=-. Similarity metrics Edge weights Weighted community detection Clusters of Input Output User interest distributions users with similar interest Figure 1. The ISCoDe framework We call ISCoDe a framewor...

Citation Context

...Girvan and Newman [16], which is highly efficient and overcomes many shortcomings of previously proposed algorithms, such as graph partitioning (e.g., spectral bisection [17], Kernighan-Lin algorithm =-=[11]-=-) and hierarchical methods (e.g., Euclidean distance single linkage clustering) [8]. These methods are not ideal for analyzing general network data since usually it is not known in advance in how many...

Citation Context

...the social network, with edge weights corresponding to the similarity metric values. In a second step, ISCoDe can invoke standard community detection algorithms for weighted graphs (for example, [14] =-=[4]-=-) to group nodes into disjoint clusters, connected internally by high-weight edges and to other subsets’ nodes with smallor zero-weight edges. These algorithms also assess, in the same time, the quali...

Citation Context

... algorithm thereof is that of Girvan and Newman [16], which is highly efficient and overcomes many shortcomings of previously proposed algorithms, such as graph partitioning (e.g., spectral bisection =-=[17]-=-, Kernighan-Lin algorithm [11]) and hierarchical methods (e.g., Euclidean distance single linkage clustering) [8]. These methods are not ideal for analyzing general network data since usually it is no...

Citation Context

... is mandatory. In many cases, a returnspath does not exist or is not efficient to provide, thus limiting thesavailable alternatives for error recovery. Packet-level ForwardsError Correction (FEC) [2]-=-=[4]-=- and data carousels [5] [6] arespossible mechanisms in this context since they can work inspush-type delivery mode without any user feedback.sThe value of the two techniques is reflected in their broa...

Citation Context

...cases, a returnspath does not exist or is not efficient to provide, thus limiting thesavailable alternatives for error recovery. Packet-level ForwardsError Correction (FEC) [2]-[4] and data carousels =-=[5]-=- [6] arespossible mechanisms in this context since they can work inspush-type delivery mode without any user feedback.sThe value of the two techniques is reflected in their broadsuse in several wirele...

Citation Context

...er separation from each other should see higher Q values than equinumerous yet “looser" structures. Moreover, with respect to ISCoDe’s resolution, we recall the remarks by Fortunato and Barthèlemy in =-=[9]-=- that algorithms seeking to maximize modularity may fail to identify important structures smaller than a scale. In concluding whether the identification of further distinct communities within a single...

Citation Context

...ortant structures smaller than a scale. In concluding whether the identification of further distinct communities within a single one is meaningful, we adopt the weak “community” condition by Radicchi =-=[18]-=-, i.e., a community c is correctly identified as one if lc L − ( dc 2L )2 > 0. (4) Note that in ISCoDe the resulting modularity values are significantly affected by the choice of the similarity metric...

Citation Context

...lesssenvironments, where transmission errors are more frequent thansin wired networks. The mobility of the users exacerbates thessituation: signal reception quality is then affected by fadingseffects =-=[1]-=-, while in cellular networks, cell handover may lead tostemporary loss of connectivity. Data loss may becomesunacceptable for the application requirements, so that somesform of error recovery is manda...

Citation Context

...n of the social network, with edge weights corresponding to the similarity metric values. In a second step, ISCoDe can invoke standard community detection algorithms for weighted graphs (for example, =-=[14]-=- [4]) to group nodes into disjoint clusters, connected internally by high-weight edges and to other subsets’ nodes with smallor zero-weight edges. These algorithms also assess, in the same time, the q...

Citation Context

... modularity in general graphs has been formulated as an Integer Linear Program (ILP) and shown to be NP-hard [2]. Proposed heuristic algorithms for modularity maximization draw on simulated annealing =-=[19]-=- or extremal optimization [6]. More commonly used and computationally friendlier, however, is the greedy agglomerative clustering algorithm of Clauset et al. [4], [15]. We simply extend it to weighted...

Citation Context

...has been formulated as an Integer Linear Program (ILP) and shown to be NP-hard [2]. Proposed heuristic algorithms for modularity maximization draw on simulated annealing [19] or extremal optimization =-=[6]-=-. More commonly used and computationally friendlier, however, is the greedy agglomerative clustering algorithm of Clauset et al. [4], [15]. We simply extend it to weighted graphs by directly relating ...

Citation Context

...aph, lc is the sum of weights over edges lying fully within community c, and dc the respective sum over the full set of edges incident to nodes in c. Modularity takes values in the interval [−1/2, 1] =-=[2]-=-. It becomes zero for community structures that do not differ than what one would get by random chance, whereas values above 0.3− 0.4 suggest strong community structure. Our framework lends to the use...

Citation Context

...a small ϵ > 0 probability. B. From weighted graphs to communities Out of the full population of clustering algorithms, relevant to our objectives are those carrying out density-based graph clustering =-=[3]-=-. Namely, they take as input a graph and partition it in a way that some notion of density (in our case: the weights of intra-cluster edges) is significantly higher within a partition than across diff...

Citation Context

...s, a returnspath does not exist or is not efficient to provide, thus limiting thesavailable alternatives for error recovery. Packet-level ForwardsError Correction (FEC) [2]-[4] and data carousels [5] =-=[6]-=- arespossible mechanisms in this context since they can work inspush-type delivery mode without any user feedback.sThe value of the two techniques is reflected in their broadsuse in several wireless s...

Citation Context

...ty and how frequently each data itemsappears in the schedule; in an optimum schedule design, the itemsfrequency of appearance in the schedule is closely related to thesitem demand probabilities { }iq =-=[14]-=-. In particular, it is shown ins[15] that under the optimum broadcast schedule design strategy,sthe average response time is given by:s( ) 2 1 21 2 1 +⋅⋅⋅= ∑ = M i iiiopt rlqS , (1)swhere ...

Citation Context

...sappears in the schedule; in an optimum schedule design, the itemsfrequency of appearance in the schedule is closely related to thesitem demand probabilities { }iq [14]. In particular, it is shown ins=-=[15]-=- that under the optimum broadcast schedule design strategy,sthe average response time is given by:s( ) 2 1 21 2 1 +⋅⋅⋅= ∑ = M i iiiopt rlqS , (1)swhere irsis the mean number of required re...

Citation Context

...very is mandatory. In many cases, a returnspath does not exist or is not efficient to provide, thus limiting thesavailable alternatives for error recovery. Packet-level ForwardsError Correction (FEC) =-=[2]-=--[4] and data carousels [5] [6] arespossible mechanisms in this context since they can work inspush-type delivery mode without any user feedback.sThe value of the two techniques is reflected in their ...

Citation Context

... previously proposed algorithms, such as graph partitioning (e.g., spectral bisection [17], Kernighan-Lin algorithm [11]) and hierarchical methods (e.g., Euclidean distance single linkage clustering) =-=[8]-=-. These methods are not ideal for analyzing general network data since usually it is not known in advance in how many communities the network should be split into and which is the best division. Newma...

Citation Context

... (e.g., interpolation and backingoff schemes) can be used. These have been studied in statistical language modeling in order to estimate the distribution of natural language as accurately as possible =-=[13]-=-. In our case non-zero request rates for interest classes can be discounted with different discounting methods (see [13]), whereas interest classes for which there is no interest can be given a small ...

Citation Context

...tection algorithm). Part of our work, hence, is devoted to the analysis and assessment of these options. For the derivation of graph edge weights, we consider two metrics: the Proportional Similarity =-=[20]-=- and the inverse of the symmetrized KullbackLeibler divergence [12]. Effectively, each metric could be seen as a different transformation from one data set (distribution of user interests over interes...

Citation Context

...r example, that higher similarity in the interests/preferences of online social group members favors collaborative, and even altruistic, behavior in content replication [10] and content dissemination =-=[1]-=- scenarios. But is such similarity present in social networks, where users tend to select their friends/followers with very different criteria, including acquaintance, social status, educational and f...

Citation Context

...ltigraph and proposed an algorithm for detecting communities in weighted networks [14]. The graph edge weights introduce another set of variables in the community detection process and it is shown in =-=[7]-=- that they can have big influence on the resulting community structure, especially on dense networks. Contribution of this paper: As in Newman’s approach, current practice in community detection consi...

Citation Context

...avior. It has been reported, for example, that higher similarity in the interests/preferences of online social group members favors collaborative, and even altruistic, behavior in content replication =-=[10]-=- and content dissemination [1] scenarios. But is such similarity present in social networks, where users tend to select their friends/followers with very different criteria, including acquaintance, so...

Citation Context

...nt-to-multipoint service over cdma2000,sthe third generation radio interface adopted in North America,senvisages the encoding of radio access layer frames beforesforwarding them to the physical layer =-=[12]-=-. Finally, bothstechniques are investigated for the design of the reliablestransport layer in the emerging Satellite Digital MultimediasBroadcasting system [13], a hybrid satellite-terrestrial systems...

Citation Context

... their broadsuse in several wireless systems and in ongoing standardizationswork. The Internet Engineering Task Force (IETF) has specifiedsa FEC building block for use in reliable multicast protocols =-=[7]-=-s[8], whilst in the European Digital Video Broadcasting (DVB)sstandard for broadcasting to mobile handheld terminals (DVB-Hsor DVB-Handheld [9]) packet-level FEC is adopted in thesMulti-Protocol Encap...

Citation Context

...ir broadsuse in several wireless systems and in ongoing standardizationswork. The Internet Engineering Task Force (IETF) has specifiedsa FEC building block for use in reliable multicast protocols [7]s=-=[8]-=-, whilst in the European Digital Video Broadcasting (DVB)sstandard for broadcasting to mobile handheld terminals (DVB-Hsor DVB-Handheld [9]) packet-level FEC is adopted in thesMulti-Protocol Encapsula...

Citation Context

...esforwarding them to the physical layer [12]. Finally, bothstechniques are investigated for the design of the reliablestransport layer in the emerging Satellite Digital MultimediasBroadcasting system =-=[13]-=-, a hybrid satellite-terrestrial systemsthat deploys a unidirectional satellite component for the deliverysof point-to-multipoint services to large, geographicallysdispersed audiences, in close synerg...