#### DMCA

## Flow-based Influence Graph Visual Summarization

### Citations

3769 | Normalized cuts and image segmentation
- Shi, Malik
- 2000
(Show Context)
Citation Context ...ard, and backward settings), one using SimRank algorithm [5] to compute the similarity matrix for SymNMF, the classical graph clustering algorithm with Ratio Association and Normalized Cut objectives =-=[6]-=-, agglomerative Modularity-based graph clustering [7], Metis Kway graph partition [8] and the Minimal Description Length (MDL) based graph summarization [9]. Note that modularity clustering is execute... |

1466 |
Finding and evaluating community structure in networks
- Newman, Girvan
(Show Context)
Citation Context ... The similar problem is also studied in the context of community detection by interdisciplinary researchers [19], in which modularity is one of the most popular quality function to access a community =-=[20]-=-. However, most of the clustering and community detection methods on graph target at maximizing intra-cluster connections while minimizing inter-cluster connections. This is fairly different from the ... |

984 | Maximizing the spread of influence through a social network
- Kempe, Kleinberg, et al.
- 2003
(Show Context)
Citation Context ...y related, IGS problem bears some subtle difference from the existing work. First (influence maximization), many elegant algorithms have been proposed for the so-called influence maximization problem =-=[1]-=-. While effective in identifying who are most influential in the graph, the question of what makes them influential largely remains open. Second (graph summarization), many interesting work has been d... |

811 | Community detection in graphs
- Fortunato
(Show Context)
Citation Context ... measures have been proposed, e.g. ratio association, ratio cut [11] and normalized cut [6]. The similar problem is also studied in the context of community detection by interdisciplinary researchers =-=[12]-=-. However, most of the clustering and community detection methods on graph target at maximizing intra-cluster connections while minimizing inter-cluster connections. This is fairly different from the ... |

691 | An algorithm for drawing general undirected graphs
- Kamada, Kawai
- 1989
(Show Context)
Citation Context ...t few decades, the methodology to draw node-link graphs has reached its maturity. On graphs with less than a few hundred nodes, the planar graph drawing approach [22] and the force-directed algorithm =-=[23]-=- can produce visually pleasant graph layouts in real time, mainly by minimizing edge crossings. On large graphs with a thousand or more nodes, the force-based algorithms can be extended by multilevel ... |

673 |
On the shortest spanning subtree of a graph and the traveling salesman problem
- Kruskal
- 1956
(Show Context)
Citation Context ...to keep a connected graph in the summarization. It is achieved by adding back the most dense flow going to each node cluster. An alternative choice is to use the maximum spanning tree (MST) algorithm =-=[4]-=-. V. EVALUATION In this section, we evaluate the proposed IGS framework and the CommonNeighbor algorithms by comparing with alternative graph summarization methods. Nine approaches (a) k = 10, l = 10 ... |

628 | Parallel multilevel k-way partitioning scheme for irregular graphs
- Karypis, Kumar
- 1996
(Show Context)
Citation Context ...y matrix for SymNMF, the classical graph clustering algorithm with Ratio Association and Normalized Cut objectives [6], agglomerative Modularity-based graph clustering [7], Metis Kway graph partition =-=[8]-=- and the Minimal Description Length (MDL) based graph summarization [9]. Note that modularity clustering is executed agglomeratively until all clusters stop merging at the top level or the number of c... |

558 |
Fast algorithm for detecting community structure in networks
- Newman
- 2004
(Show Context)
Citation Context ...thm [5] to compute the similarity matrix for SymNMF, the classical graph clustering algorithm with Ratio Association and Normalized Cut objectives [6], agglomerative Modularity-based graph clustering =-=[7]-=-, Metis Kway graph partition [8] and the Minimal Description Length (MDL) based graph summarization [9]. Note that modularity clustering is executed agglomeratively until all clusters stop merging at ... |

384 | Simrank: A measure of structural-context similarity
- Jeh, Widom
(Show Context)
Citation Context ...ween k clusters. are considered: three using CommonNeighbor algorithms to compute the similarity matrix for SymNMF (i.e. forward+backward, forward, and backward settings), one using SimRank algorithm =-=[5]-=- to compute the similarity matrix for SymNMF, the classical graph clustering algorithm with Ratio Association and Normalized Cut objectives [6], agglomerative Modularity-based graph clustering [7], Me... |

245 |
The structural equivalence of individuals in social networks
- Lorrain, White
- 1971
(Show Context)
Citation Context ...ion was proposed to present the graph with an aggregated structure and an error correction list. On influence graphs which are sparse, it performs similarly to a structural equivalence based grouping =-=[13]-=-, leaving huge visual clutters unsettled. Meanwhile, Shahaf et al. [14][15] studied the similar problem of summarizing large amount of information into user-friendly visual maps. On a quite different ... |

200 | Arnetminer: Extraction and mining of academic social networks
- Tang, Zhang, et al.
- 2008
(Show Context)
Citation Context ...ead to a much higher overall flow rate), we exclude MDL from numeric comparisons, but present its visual summarization results. The experiment data are paper citation graphs collected from ArnetMiner =-=[10]-=-. The influence graphs are obtained by reversing citation links. A. Flow Rate Maximization We first pick five source papers from the data set to generate maximal influence graphs. These influence grap... |

196 | Spectral relaxation for k-means clustering
- Zha, Ding, et al.
(Show Context)
Citation Context ...2|pic(s)||pid(s)| given {pic} k c=1 (7) The second part of the optimization can be solved by selecting l top flows with the largest size-normalized flow rate. B. Kernel K-Mean Clustering According to =-=[3]-=-, the kernel k-mean clustering (KM) is defined as follows. Given n data vectors {xi}ni=1 with kernel function φ(xi), KM method groups the data vectors into k nonoverlapping clusters {pic}kc=1 based on... |

173 |
Spectral k-way ratio cut partitioning and clustering
- Chan, Schlag, et al.
- 1994
(Show Context)
Citation Context ...ph clustering algorithms. These algorithms usually optimize certain association or cut measure during the k-way graph partition. Several measures have been proposed, e.g. ratio association, ratio cut =-=[11]-=- and normalized cut [6]. The similar problem is also studied in the context of community detection by interdisciplinary researchers [12]. However, most of the clustering and community detection method... |

160 |
On the readability of graphs using node-link and matrix-based representations: a controlled experiment and statistical analysis
- Ghoniem, Fekete, et al.
(Show Context)
Citation Context ...ual complexity doubles. We recommend to set k ≤ 20 on this trade-off, because at a size larger than 20, the node-link graph visualization may not be a good choice for many graph visual analysis tasks =-=[16]-=-. On the choice of l, we find that the optimization of IGS objective is not significant after l ≥ 2k, therefore l = 2k can be an appropriate setting. Last, we target on the academic data sets in this ... |

151 | On the equivalence of nonnegative matrix factorization and spectral clustering
- Ding, He, et al.
(Show Context)
Citation Context ... propose a matrix decomposition based solution to generate k node clusters from the similarity matrix MG. The decomposition employs a Symmetric version of the Nonnegative Matrix Factorization (SymNMF =-=[2]-=-) which optimizes: min H≥0 ||MG −HHT ||2F (3) where ||·||F denotes the Frobenius norm of the matrix. H = {hij} is a n by k matrix indicating the cluster membership 1. Pick source node f 2. Rooted sear... |

150 | Social influence analysis in large-scale networks
- Tang, Sun, et al.
- 2009
(Show Context)
Citation Context ...ng the effects of social influence. For example, Bakshy et al. [16] conducted randomized controlled trials to identify the effect of social influence on consumer responses to advertising. Tang et al. =-=[17]-=- presented a Topical Affinity Propagation (TAP) approach to quantify the topic-level social influence in large networks. Kempe et al. [1] proposed to use a submodular function to formalize the influen... |

146 |
Graph Drawing: Algorithms for the Visualization of Graphs
- Battista, Eades, et al.
- 1999
(Show Context)
Citation Context ...o. Second, visualization, over the past few decades, the methodology to draw node-link graphs has reached its maturity. On graphs with less than a few hundred nodes, the planar graph drawing approach =-=[22]-=- and the force-directed algorithm [23] can produce visually pleasant graph layouts in real time, mainly by minimizing edge crossings. On large graphs with a thousand or more nodes, the force-based alg... |

83 | Efficient aggregation for graph summarization
- Tian, Hankins, et al.
- 2008
(Show Context)
Citation Context ...ser (the citation graphs have an average degree of less than 3), it performs similarly to a structural equivalence based grouping [17], leaving huge visual clutters unsettled. Another algorithm, SNAP =-=[21]-=-, considers the node attribute on graph, but again is not tailored for the influence graph scenario. Second, visualization, over the past few decades, the methodology to draw node-link graphs has reac... |

78 | Graph summarization with bounded error
- Navlakha, Rastogi, et al.
- 2008
(Show Context)
Citation Context ...o Association and Normalized Cut objectives [6], agglomerative Modularity-based graph clustering [7], Metis Kway graph partition [8] and the Minimal Description Length (MDL) based graph summarization =-=[9]-=-. Note that modularity clustering is executed agglomeratively until all clusters stop merging at the top level or the number of clusters reaches k. The MDL algorithm can not specify the number of clus... |

67 |
A 61-million-person experiment in social influence and political mobilization
- Bond, Fariss, et al.
- 2012
(Show Context)
Citation Context ...ng the effects of social influence. For example, Bakshy et al. [28] conducted randomized controlled trials to identify the effect of social influence on consumer responses to advertising. Bond et al. =-=[29]-=- used a randomized controlled trial to verify the social influence on political voting behavior. Tang et al. [30] presented a Topical Affinity Propagation (TAP) approach to quantify the topic-level so... |

54 | SVD based initialization: a head start for nonnegative matrix factorization
- Boutsidis, Gallopoulos
- 2008
(Show Context)
Citation Context ...e in [2] which guarantees convergence. In this iterative algorithm, the initialization of H is critical to the final result. We introduce nonnegative eigenvalue decomposition similar to the method in =-=[3]-=- to compute a good initial factorization. Link Pruning. The graph summarization by SymNMF needs further post-processing to select l top flows for the final summarization S. Here we extract the top flo... |

48 | Connecting the dots between news articles
- Shahaf, Guestrin
- 2010
(Show Context)
Citation Context ...rror correction list. On influence graphs which are sparse, it performs similarly to a structural equivalence based grouping [13], leaving huge visual clutters unsettled. Meanwhile, Shahaf et al. [14]=-=[15]-=- studied the similar problem of summarizing large amount of information into user-friendly visual maps. On a quite different focus, our method is built on the graph with explicit linkage data while th... |

33 |
Social influence in social advertising: Evidence from field experiments
- Bakshy, Eckles, et al.
- 2012
(Show Context)
Citation Context ... linkage data while the textual content of each node can be absent or incomplete. Second, considerable work has been conducted for studying the effects of social influence. For example, Bakshy et al. =-=[16]-=- conducted randomized controlled trials to identify the effect of social influence on consumer responses to advertising. Tang et al. [17] presented a Topical Affinity Propagation (TAP) approach to qua... |

25 |
Efficient and high quality force-directed graph drawing.
- Hu
- 2005
(Show Context)
Citation Context ...outs in real time, mainly by minimizing edge crossings. On large graphs with a thousand or more nodes, the force-based algorithms can be extended by multilevel coarsening and fast force approximation =-=[24]-=- and still generate a layout in reasonable time (e.g. less than a minute for million-node graphs). However, on real-world large graphs with small-world nature, including the influence graph discussed ... |

24 | Trains of thought: Generating information maps
- Shahaf, Guestrin, et al.
- 2012
(Show Context)
Citation Context ...ce graph discussed here, the resulting graph layout still has numerous edge crossings. This leads to overwhelming visual clutters detrimental to visual data mining tasks. Meanwhile, Shahaf et al. [25]=-=[26]-=-[27] studied the similar problem of summarizing large amount of information into userfriendly visual maps. They developed intriguing methods to detect hidden linkage and document clusters from the key... |

20 | H (2012) Symmetric nonnegative matrix factorization for graph clustering
- Kuang, Ding, et al.
(Show Context)
Citation Context ...and the Minimal Description Length (MDL) based graph summarization [12]. Note that Ratio Association and Normalized Cut are implemented using their equivalent similarity matrix computation for SymNMF =-=[13]-=-. Metis partition is implemented by official open source software package [14]. Modularity clustering is executed agglomeratively until all clusters stop merging at the top level or the number of clus... |

14 | Cascade-based community detection
- Barbieri, Bonchi, et al.
- 2013
(Show Context)
Citation Context ...n problem. Recently, Mehmood et al. proposed CSI [31], a model that generalizes the classical Independent Cascade model to the community level, built from the cascade-based community detection method =-=[32]-=-. CSI can produce similar visual forms to our result. However, the CSI model is computed from the probabilistic social influence graph and the information propagation log more engaged to the social in... |

13 | Information cartography: Creating zoomable, large-scale maps of information
- Shahaf, Yang, et al.
- 2013
(Show Context)
Citation Context ...an error correction list. On influence graphs which are sparse, it performs similarly to a structural equivalence based grouping [13], leaving huge visual clutters unsettled. Meanwhile, Shahaf et al. =-=[14]-=-[15] studied the similar problem of summarizing large amount of information into user-friendly visual maps. On a quite different focus, our method is built on the graph with explicit linkage data whil... |

3 |
Csi: Community-level social influence analysis
- Mehmood, Barbieri, et al.
(Show Context)
Citation Context ... Most of these works focus on the existence of social influence or the nature of the information diffusion process and do not consider the summarization problem. Recently, Mehmood et al. proposed CSI =-=[18]-=-, a model that generalizes the classical Independent Cascade model to the community level. CSI can produce similar visual forms to our result. However, the CSI model is designed for the social influen... |