#### DMCA

## Dynamic behavioral mixedmembership model for large evolving networks (2012)

Venue: | In Arxiv preprint arXiv:1205.2056 |

Citations: | 1 - 1 self |

### Citations

534 | Graphs over time: densification laws, shrinking diameters and possible explanations
- Leskovec, Kleinberg, et al.
- 2005
(Show Context)
Citation Context ...re systems to optimally manage data flow, to detect fraud and intrusions, and to allocate resources for growth over time. Although some recent research has focused on the analysis of dynamic networks =-=[17,3,5,12,4,20]-=-, there has been less work on developing models of temporal behavior in large scale network datasets. There has been some work on modeling temporal events in large scale networks [2,28] and other work... |

489 | Group formation in large social networks: Membership, growth, and evolution
- Backstrom, Huttenlocher, et al.
- 2006
(Show Context)
Citation Context ...re systems to optimally manage data flow, to detect fraud and intrusions, and to allocate resources for growth over time. Although some recent research has focused on the analysis of dynamic networks =-=[17,3,5,12,4,20]-=-, there has been less work on developing models of temporal behavior in large scale network datasets. There has been some work on modeling temporal events in large scale networks [2,28] and other work... |

451 | The dynamics of viral marketing
- Leskovec, Adamic, et al.
- 2006
(Show Context)
Citation Context ...tive for real-time analysis of large streaming graphs. 4 Related Work There has been an abundance of work in analyzing dynamic networks. However, the majority of this work focuses on dynamic patterns =-=[10,17,16,21,25]-=-, tempoAlgorithm 1 Anomalous Structural Transitions Input: G = {Gt : t = 1, ..., tmax} (evolving mixed-memberships) Output: x (vector of anomalous scores) 1: for i = 1→ n do 2: T (i) ∈ Rr×r ← NMF (G(i... |

187 |
A simple generalisation of the area under the roc curve for multiple class classification problems.
- Hand, Till
- 2001
(Show Context)
Citation Context ...ship). The predicted class label for node i is the modal role from the ith row of Ĝt+1. We evaluate the predictions using a generalization of AUC extended for multi-class problems (a.k.a. Total AUC) =-=[13]-=-. Fig. 2 shows that the DBMM summary transition model is an effective predictor across the range of experiments. With few exceptions, our model outperforms both baselines for all data sets and timeste... |

154 | Graphscope: parameter-free mining of large time-evolving graphs
- Sun, Faloutsos, et al.
- 2007
(Show Context)
Citation Context ...improve predictive models [24]. In addition, there is work on ar X iv :1 20 5. 20 56 v1s[ cs .SI ]s9 M ays20 12 2 R. Rossi, B. Gallagher, J. Neville, K. Henderson identifying clusters in dynamic data =-=[5,25]-=- but these methods focus on discovering underlying communities over time—sets of nodes that are highly clustered together. In contrast, we are interested in uncovering the behavioral patterns of nodes... |

144 | Patterns of temporal variation in online media - Yang, Leskovec - 2011 |

131 |
Evolutionary clustering
- Chakrabarti, Kumar, et al.
- 2006
(Show Context)
Citation Context ...re systems to optimally manage data flow, to detect fraud and intrusions, and to allocate resources for growth over time. Although some recent research has focused on the analysis of dynamic networks =-=[17,3,5,12,4,20]-=-, there has been less work on developing models of temporal behavior in large scale network datasets. There has been some work on modeling temporal events in large scale networks [2,28] and other work... |

104 | Streaming pattern discovery in multiple time-series
- Papadimitriou, Sun, et al.
- 2005
(Show Context)
Citation Context ...tive for real-time analysis of large streaming graphs. 4 Related Work There has been an abundance of work in analyzing dynamic networks. However, the majority of this work focuses on dynamic patterns =-=[10,17,16,21,25]-=-, tempoAlgorithm 1 Anomalous Structural Transitions Input: G = {Gt : t = 1, ..., tmax} (evolving mixed-memberships) Output: x (vector of anomalous scores) 1: for i = 1→ n do 2: T (i) ∈ Rr×r ← NMF (G(i... |

94 | An event-based framework for characterizing the evolutionary behavior of interaction graphs
- Asur, Parthasarathy, et al.
- 2007
(Show Context)
Citation Context ...orks [17,3,5,12,4,20], there has been less work on developing models of temporal behavior in large scale network datasets. There has been some work on modeling temporal events in large scale networks =-=[2,28]-=- and other work that uses temporal link patterns to improve predictive models [24]. In addition, there is work on ar X iv :1 20 5. 20 56 v1s[ cs .SI ]s9 M ays20 12 2 R. Rossi, B. Gallagher, J. Neville... |

62 | Community evolution in dynamic multi-mode networks
- Tang, Liu, et al.
- 2008
(Show Context)
Citation Context ... NMF (G(i)1:t−1,G(i)2:t) 3: Ĝt+1 = T (i) ·Gt 4: x(i) = ∥∥∥Ĝt+1 −Gt+1∥∥∥ F 5: end for Dynamic Behavioral Mixed-Membership Model 15 ral link prediction [8], anomaly detection [1], dynamic communities =-=[18,26,11]-=-, dynamic node ranking [19,23], and many others [28,12]. In contrast, we propose a scalable dynamic mixed-membership model that captures the node behaviors over time and consequently learns a predicti... |

61 | Ten years in the evolution of the Internet ecosystem
- Dhamdhere, Dovrolis
- 2008
(Show Context)
Citation Context ...anding that the underlying evolutionary process of the Internet AS is not stationary and matches recent evidence of the Internet topology transitioning from hierarchal to a flat topological structure =-=[7,6]-=-. Furthermore, the spike in loss (e.g., Fig. 2(h)) provide insights into network-level anomalies which could be due to large-scale emergencies, holiday seasons, or other major events. In the larger Tw... |

50 | The Internet is Flat: Modeling the transition from a transit hierarchy to a peering mesh
- Dhamdhere, Dovrolis
- 2010
(Show Context)
Citation Context ...anding that the underlying evolutionary process of the Internet AS is not stationary and matches recent evidence of the Internet topology transitioning from hierarchal to a flat topological structure =-=[7,6]-=-. Furthermore, the spike in loss (e.g., Fig. 2(h)) provide insights into network-level anomalies which could be due to large-scale emergencies, holiday seasons, or other major events. In the larger Tw... |

48 | Analyzing Communities and Their Evolutions in Dynamic Social Networks.
- Lin, Chi, et al.
- 2009
(Show Context)
Citation Context ... NMF (G(i)1:t−1,G(i)2:t) 3: Ĝt+1 = T (i) ·Gt 4: x(i) = ∥∥∥Ĝt+1 −Gt+1∥∥∥ F 5: end for Dynamic Behavioral Mixed-Membership Model 15 ral link prediction [8], anomaly detection [1], dynamic communities =-=[18,26,11]-=-, dynamic node ranking [19,23], and many others [28,12]. In contrast, we propose a scalable dynamic mixed-membership model that captures the node behaviors over time and consequently learns a predicti... |

47 |
Prediction and ranking algorithms for event-based network data
- O’Madadhain, Hutchins, et al.
- 2005
(Show Context)
Citation Context ... = T (i) ·Gt 4: x(i) = ∥∥∥Ĝt+1 −Gt+1∥∥∥ F 5: end for Dynamic Behavioral Mixed-Membership Model 15 ral link prediction [8], anomaly detection [1], dynamic communities [18,26,11], dynamic node ranking =-=[19,23]-=-, and many others [28,12]. In contrast, we propose a scalable dynamic mixed-membership model that captures the node behaviors over time and consequently learns a predictive model for how these behavio... |

45 |
Temporal link prediction using matrix and tensor factorizations,”
- Dunlavy, Kolda, et al.
- 2011
(Show Context)
Citation Context ...ous scores) 1: for i = 1→ n do 2: T (i) ∈ Rr×r ← NMF (G(i)1:t−1,G(i)2:t) 3: Ĝt+1 = T (i) ·Gt 4: x(i) = ∥∥∥Ĝt+1 −Gt+1∥∥∥ F 5: end for Dynamic Behavioral Mixed-Membership Model 15 ral link prediction =-=[8]-=-, anomaly detection [1], dynamic communities [18,26,11], dynamic node ranking [19,23], and many others [28,12]. In contrast, we propose a scalable dynamic mixed-membership model that captures the node... |

40 | Tracking the evolution of communities in dynamic social networks.
- Greene, Doyle, et al.
- 2010
(Show Context)
Citation Context ... NMF (G(i)1:t−1,G(i)2:t) 3: Ĝt+1 = T (i) ·Gt 4: x(i) = ∥∥∥Ĝt+1 −Gt+1∥∥∥ F 5: end for Dynamic Behavioral Mixed-Membership Model 15 ral link prediction [8], anomaly detection [1], dynamic communities =-=[18,26,11]-=-, dynamic node ranking [19,23], and many others [28,12]. In contrast, we propose a scalable dynamic mixed-membership model that captures the node behaviors over time and consequently learns a predicti... |

38 | Dynamic mixed membership blockmodel for evolving networks
- Fu, Song, et al.
(Show Context)
Citation Context ... interested in uncovering the behavioral patterns of nodes in the network and modeling how those patterns change over time. The recent work on dynamic mixed-membership stochastic block models (dMMSB: =-=[9,27]-=-), is to our knowledge, one of the only methods suitable for modeling nodecentric behavior over time. The dMMSB model identifies groups of nodes with similar patterns of linkage and characterizes how ... |

37 | A state-space mixed membership blockmodel for dynamic network tomography.
- Xing, Fu, et al.
- 2010
(Show Context)
Citation Context ... interested in uncovering the behavioral patterns of nodes in the network and modeling how those patterns change over time. The recent work on dynamic mixed-membership stochastic block models (dMMSB: =-=[9,27]-=-), is to our knowledge, one of the only methods suitable for modeling nodecentric behavior over time. The dMMSB model identifies groups of nodes with similar patterns of linkage and characterizes how ... |

31 |
Measurement and Analysis of
- Mislove
- 2007
(Show Context)
Citation Context |

27 | It’s who you know: graph mining using recursive structural features
- Henderson, Gallagher, et al.
(Show Context)
Citation Context ...re extracted via a two-step process. Feature Discovery. The first step is to represent each active node in a given snapshot graph St using a set of representative features. For this task, we leverage =-=[14]-=-. The method constructs degree and egonet measures (in/out, weighted,...), then aggregates these measures using sum/mean creating recursive features. After each aggregation step, correlated features a... |

26 | Finding spread blockers in dynamic networks
- Habiba, Yu, et al.
- 2008
(Show Context)
Citation Context |

19 |
Path lengths, correlations, and centrality in temporal networks. arXiv,
- Pan, Saramaki
- 2011
(Show Context)
Citation Context |

15 | C.: Modeling blog dynamics
- Götz, Leskovec, et al.
- 2009
(Show Context)
Citation Context ...tive for real-time analysis of large streaming graphs. 4 Related Work There has been an abundance of work in analyzing dynamic networks. However, the majority of this work focuses on dynamic patterns =-=[10,17,16,21,25]-=-, tempoAlgorithm 1 Anomalous Structural Transitions Input: G = {Gt : t = 1, ..., tmax} (evolving mixed-memberships) Output: x (vector of anomalous scores) 1: for i = 1→ n do 2: T (i) ∈ Rr×r ← NMF (G(i... |

8 |
Detecting novel discrepancies in communication networks
- Abello, Eliassi-Rad, et al.
- 2010
(Show Context)
Citation Context ...1→ n do 2: T (i) ∈ Rr×r ← NMF (G(i)1:t−1,G(i)2:t) 3: Ĝt+1 = T (i) ·Gt 4: x(i) = ∥∥∥Ĝt+1 −Gt+1∥∥∥ F 5: end for Dynamic Behavioral Mixed-Membership Model 15 ral link prediction [8], anomaly detection =-=[1]-=-, dynamic communities [18,26,11], dynamic node ranking [19,23], and many others [28,12]. In contrast, we propose a scalable dynamic mixed-membership model that captures the node behaviors over time an... |

7 | Time-evolving relational classification and ensemble methods
- Rossi, Neville
- 2012
(Show Context)
Citation Context ...avior in large scale network datasets. There has been some work on modeling temporal events in large scale networks [2,28] and other work that uses temporal link patterns to improve predictive models =-=[24]-=-. In addition, there is work on ar X iv :1 20 5. 20 56 v1s[ cs .SI ]s9 M ays20 12 2 R. Rossi, B. Gallagher, J. Neville, K. Henderson identifying clusters in dynamic data [5,25] but these methods focus... |

4 |
Modeling temporal behavior in large networks: A dynamic mixed-membership model
- Rossi, Gallagher, et al.
- 2011
(Show Context)
Citation Context ... network evolution, and the type of time-evolving network (e.g., transactional vs. social network). Some results, including the AUC plots are omitted for space, but all are qualitatively similar, see =-=[22]-=-. In addition to demonstrating that our model is an effective predictor, Fig. 2 offers some interesting insights into the underlying dynamics of these networks. For instance, the drift we see in Fig. ... |

3 | RolX: Role Extraction and Mining in Large Networks
- Henderson, Gallagher, et al.
- 2011
(Show Context)
Citation Context ...step is to automatically discover groups of nodes (representing common patterns of behavior) based on their features. For this purpose, we use Non-negative Matrix Factorization (NMF) to extract roles =-=[15]-=- and extend it for a sequence of graphs. Given a sequence of node-feature matrices, we generate a rank-r approximationGtF ≈ V t where each row ofGt ∈ Rn×r represents a node’s membership in each role a... |

3 | Dynamic pagerank using evolving teleportation
- Rossi, Gleich
- 2012
(Show Context)
Citation Context ... = T (i) ·Gt 4: x(i) = ∥∥∥Ĝt+1 −Gt+1∥∥∥ F 5: end for Dynamic Behavioral Mixed-Membership Model 15 ral link prediction [8], anomaly detection [1], dynamic communities [18,26,11], dynamic node ranking =-=[19,23]-=-, and many others [28,12]. In contrast, we propose a scalable dynamic mixed-membership model that captures the node behaviors over time and consequently learns a predictive model for how these behavio... |