## TopoLayout: Multi-level graph layout by topological features (2007)

Venue: | IEEE TRANS. VISUALIZATION AND COMPUTER GRAPHICS |

Citations: | 36 - 5 self |

### BibTeX

@ARTICLE{Archambault07topolayout:multi-level,

author = {Daniel Archambault and Tamara Munzner and David Auber},

title = {TopoLayout: Multi-level graph layout by topological features},

journal = {IEEE TRANS. VISUALIZATION AND COMPUTER GRAPHICS},

year = {2007},

volume = {13},

pages = {2007}

}

### OpenURL

### Abstract

We describe TopoLayout, a feature-based, multi-level algorithm that draws undirected graphs based on the topological features they contain. Topological features are detected recursively inside the graph, and their subgraphs are collapsed into single nodes, forming a graph hierarchy. Each feature is drawn with an algorithm tuned for its topology. As would be expected from a feature-based approach, the runtime and visual quality of TopoLayout depends on the number and types of topological features present in the graph. We show experimental results comparing speed and visual quality for TopoLayout against four other multi-level algorithms on a variety of datasets with a range of connectivities and sizes. TopoLayout frequently improves the results in terms of speed and visual quality on these datasets.

### Citations

540 | A heuristic for graph drawing
- Eades
- 1984
(Show Context)
Citation Context ...n between them as a straight line. The problem of drawing general, undirected graphs has been well studied. Before the late 1990s, the methods were primarily focused on force-directed approaches [7], =-=[10]-=-–[12], [20]. These methods perform well for many types of graphs, but do not scale to graphs of thousands of nodes. To overcome this limitation, multi-level approaches and approaches which rely more h... |

511 | An Algorithm for Drawing General Undirected Graphs
- Kamada, Kawai
- 1989
(Show Context)
Citation Context ...hem as a straight line. The problem of drawing general, undirected graphs has been well studied. Before the late 1990s, the methods were primarily focused on force-directed approaches [7], [10]–[12], =-=[20]-=-. These methods perform well for many types of graphs, but do not scale to graphs of thousands of nodes. To overcome this limitation, multi-level approaches and approaches which rely more heavily on u... |

468 | Graph drawing by force-directed placement
- Fruchterman, Reingold
- 1991
(Show Context)
Citation Context ...ween them as a straight line. The problem of drawing general, undirected graphs has been well studied. Before the late 1990s, the methods were primarily focused on force-directed approaches [7], [10]–=-=[12]-=-, [20]. These methods perform well for many types of graphs, but do not scale to graphs of thousands of nodes. To overcome this limitation, multi-level approaches and approaches which rely more heavil... |

190 | Drawing Graphs Nicely using Simulated Annealing
- Davidson, Harel
- 1996
(Show Context)
Citation Context ... drawn between them as a straight line. The problem of drawing general, undirected graphs has been well studied. Before the late 1990s, the methods were primarily focused on force-directed approaches =-=[7]-=-, [10]–[12], [20]. These methods perform well for many types of graphs, but do not scale to graphs of thousands of nodes. To overcome this limitation, multi-level approaches and approaches which rely ... |

155 |
Computer Algorithms: Introduction to Design and Analysis
- Baase, AV
- 2001
(Show Context)
Citation Context ...nnot be found. 3) Connected Components: We detect connected components using a series of depth-first searches to compute spanning trees for each component. We refer the reader to Baase and Van Gelder =-=[5]-=- for details of this standard algorithm which runs in O(N + E) time. 4) Trees: We detect trees by finding the first cycle in the graph and selecting a node n on that cycle. If a cycle is not found, th... |

112 | A fast adaptive layout algorithm for undirected graphs
- Frick, Ludwig, et al.
- 1994
(Show Context)
Citation Context ...n. In the decomposition phase, we have not found previous work describing our tree detection algorithm. In the feature layout phase, we provide a weighting scheme for HDE [22] and slightly modify GEM =-=[11]-=- so that they are area-aware. The crossing reduction phase is new to multi-level algorithms and is a novel algorithm of our own. The overlap elimination phase is new to multi-level algorithms, but is ... |

96 | A Multilevel Algorithm for Force-Directed Graph Drawing
- Walshaw
- 1984
(Show Context)
Citation Context ...aid out. Also, such decompositions help avoid local minima, allowing the algorithms to scale to larger datasets, improving both the running time and the visual quality of the final layout. In Walshaw =-=[30]-=-, an estimate of a solution of the maximal matching problem is used as a coarsening operator to construct the hierarchy. The maximal matching problem is to select the largest possible set of edges in ... |

95 | Texture Mapping Using Surface Flattening via Multidimensional Scaling
- ZIGELMAN, KIMMEL, et al.
(Show Context)
Citation Context ...oice. HDE is related to a rich family of mathematical approaches which have been explored as solutions to problems ranging from flattening curved surfaces [26] to texture mapping in computer graphics =-=[31]-=-. These algorithms select a subset of d points called pivots and compute the pairwise geodesic or graph theoretic distance between the pivots and all other points on the surface. Each pivot correspond... |

86 |
Linear Algebra and Its Applications
- Lay
- 1994
(Show Context)
Citation Context ... suitable layout algorithm for the subgraph, we analyze the eigenvalues produced by an HDE layout of the graph. In PCA, the amount of variance in the data captured by an eigenvector is its eigenvalue =-=[23]-=-. We first determine if there is enough variance in the data, or if its largest eigenvalue is above a minimum threshold value. In our system, a value of 100 was determined empirically. This minimum va... |

82 | A fast multi-scale method for drawing large graph
- Koren, Harel
- 2002
(Show Context)
Citation Context ...ing operator to construct the hierarchy. The maximal matching problem is to select the largest possible set of edges in the graph such that no two edges are incident to the same node. Harel and Koren =-=[19]-=- recursively apply an approximate solution to the k-centres problem, using graph theoretic distance as the ideal distance between two nodes. The k-centres problem groups a set of points into k cluster... |

67 |
A numerical solution to the generalized mapmaker's problem: Flattening nonconvex polyhedral surfaces
- Schwartz, Shaw, et al.
- 1989
(Show Context)
Citation Context ... [22] of Harel and Koren is an appropriate choice. HDE is related to a rich family of mathematical approaches which have been explored as solutions to problems ranging from flattening curved surfaces =-=[26]-=- to texture mapping in computer graphics [31]. These algorithms select a subset of d points called pivots and compute the pairwise geodesic or graph theoretic distance between the pivots and all other... |

65 |
Multiscale visualization of small world networks
- Auber
- 2003
(Show Context)
Citation Context ...h node is connected to all others. Nodes and edges of complete graphs are coloured cyan. A cluster is a subgraph formed by some clustering algorithm. In our implementation, we use the strength metric =-=[4]-=- for clustering. Edges separating clusters are coloured grey. We then determine if HDE is a suitable algorithm to lay out the subgraph. If it is, it is coloured purple. Finally, if the decomposition p... |

62 | Graph drawing by high dimensional embedding
- Harel, Koren
- 2002
(Show Context)
Citation Context ... {archam, tmm}@cs.ubc.ca D. Auber is with University of Bordeaux I, auber@labri.fr that we can expand our system beyond strict topological features, we detect when the high-dimensional embedder (HDE) =-=[22]-=- algorithm is a suitable choice for layout. HDE is an efficient algorithm for drawing a specific subset of general graphs, many of which are grids. The primary contribution of this work is TopoLayout,... |

50 |
Interactive visualization of small world graphs
- Ham, Wijk
(Show Context)
Citation Context ...of Abello et al. [1] constructs hierarchies based on a binary space partition of the layout or areas of relatively high connectivity detected using Markov clustering. The work of van Ham and van Wijk =-=[29]-=- provides a similar technique for visualizing small world graphs by merging clusters pairwise based on geometric distance between the clusters. A force-directed layout algorithm that reflects the unde... |

46 | Topological fisheye views for visualizing large graphs
- Gansner
(Show Context)
Citation Context ...s shown at its maximum level of detail. Coarser levels of the hierarchy are displayed at increasing distances from the focus region, providing context. The topological fisheye views of Gansner et al. =-=[14]-=- constructs the hierarchies based on Delaunay triangulations and relative neighborhood graphs. The compound fisheye views of Abello et al. [1] constructs hierarchies based on a binary space partition ... |

38 | GRIP: Graph dRawing with Intelligent Placement
- Gajer, Kobourov
- 2000
(Show Context)
Citation Context ...nce as the ideal distance between two nodes. The k-centres problem groups a set of points into k clusters where the distance between any pair of points in the cluster is minimized. The GRIP algorithm =-=[13]-=- coarsens by applying a filtration to the node set of the input graph. The filtration operator recursively constructs a maximal subset at each level i such that the graph theoretic distance between an... |

35 |
Drawing large graphs with a potential-field-based multilevel algorithm
- Hachul, Jnger
- 2004
(Show Context)
Citation Context ...tion is recursively used as an estimate for the eigenvectors one level down until the eigenvectors of the original matrix have been computed. The Fast Multipole Multilevel Method, or FM 3 , algorithm =-=[16]-=- is the first multi-level algorithm for general graphs with a provable worst case asymptotic runtime of O(N logN +E). In this approach, the graph is partitioned into subgraphs called solar systems. Th... |

35 | Cluster busting in anchored graph drawing
- Lyons
- 1992
(Show Context)
Citation Context ...as slow, it does guarantee a drawing free of node-node overlaps and produced drawings of high visual quality for many types of graphs. We also implemented the Cluster Buster algorithm of Lyons et al. =-=[24]-=-, which computes the Voronoi diagram of the node set and iteratively pulls the nodes towards the centroid of each Voronoi cell. For a constant number of iterations, the algorithm runs in O(N logN) tim... |

31 | Drawing graphs with nonuniform vertices - Harel, Koren - 2002 |

30 | Drawing huge graphs by algebraic multigrid optimization, Multiscale Model
- Koren, Carmel, et al.
(Show Context)
Citation Context ...ph. The filtration operator recursively constructs a maximal subset at each level i such that the graph theoretic distance between any two nodes of the subset is at least 2 i−1 − 1. The ACE algorithm =-=[21]-=- solves for the eigenvectors of the Laplacian matrix to determine a suitable projection of the graph into two, three, or any dimension less than or equal to the number of eigenvectors of the matrix. T... |

28 | Fast node overlap removal
- Dwyer, Marriott, et al.
- 2005
(Show Context)
Citation Context ... overlaps in the final drawing, and the results were usually of low visual quality. We obtained the best results from implementing the fast node overlap removal algorithm without Lagrange multipliers =-=[8]-=-, which is discussed in detail in Dwyer et al’s technical report [9]. In this work, two separate passes along the x-axis and the y-axis eliminate all node overlaps in the graph. The algorithm construc... |

27 |
Tulip — A huge graph visualization framework
- Auber
- 2004
(Show Context)
Citation Context ...nents, for each hierarchical level. hierarchy to compute the level i positions of no and nc when computing torques. V. EMPIRICAL EVALUATION We implemented the TopoLayout framework on top of the Tulip =-=[3]-=- graph visualization system and have tested it against other multi-level algorithms on datasets with a range of connectivities and sizes. All benchmarks were run on a 3.0GHz Pentium IV with 3.0GB of m... |

23 | An experimental comparison of fast algorithms for drawing general large graphs
- Hachul, Junger
(Show Context)
Citation Context ... to prove that the final graph layout can be obtained in O(N logN +E) time. A subsequent evaluation of FM 3 convincingly demonstrates that FM 3 yields higher visual quality results than previous work =-=[17]-=-. All the multi-level algorithms described above use heuristics to construct their graph hierarchies which do not exploit low-level and high-level features in the data. The principal advantage of a fe... |

20 | Improving Walker’s Algorithm to Run in Linear Time
- Buchheim, Jünger, et al.
- 2002
(Show Context)
Citation Context ... use any tree layout algorithm that is area-aware for drawing. We use the bubble tree algorithm [15] for trees of low depth and high branching factor and an area-aware version of the Walker algorithm =-=[6]-=- for all other trees. The bubble tree algorithm requires O(N logN) time while the version of the Walker algorithm runs in O(N) time. 2) Area-Aware Circular Layout: These algorithms highlight complete ... |

15 |
Bubble Tree Drawing Algorithm
- Grivet, Auber, et al.
- 2004
(Show Context)
Citation Context ...of the nodes at lower levels in the hierarchy that connect the children to the meta-node. TopoLayout can use any tree layout algorithm that is area-aware for drawing. We use the bubble tree algorithm =-=[15]-=- for trees of low depth and high branching factor and an area-aware version of the Walker algorithm [6] for all other trees. The bubble tree algorithm requires O(N logN) time while the version of the ... |

12 |
A Framework for Circular Drawings of Networks
- Six, Tollis
- 1999
(Show Context)
Citation Context ...e visually convincing results, the largest graph drawn was a thousand nodes. No explicit performance numbers were given, but the time required for precomputation is a major limitation. Six and Tollis =-=[27]-=- decompose the graph into biconnected graphs and draw the tree of biconnected components using a radial tree layout algorithm which is area-aware. The individual biconnected components are drawn using... |

11 | Visualizing large graphs with compound-fisheye views and treemaps
- Abello, Kobourov, et al.
- 2004
(Show Context)
Citation Context ...context. The topological fisheye views of Gansner et al. [14] constructs the hierarchies based on Delaunay triangulations and relative neighborhood graphs. The compound fisheye views of Abello et al. =-=[1]-=- constructs hierarchies based on a binary space partition of the layout or areas of relatively high connectivity detected using Markov clustering. The work of van Ham and van Wijk [29] provides a simi... |

7 |
Visualization of biological information with circular drawings
- Symeonidis, Tollis
- 2004
(Show Context)
Citation Context ...an elimination of node-edge or edge-edge crossings, but it reduces their number in most cases and also shortens edge length between subgraphs. Our approach is similar to that of Symeonidis and Tollis =-=[28]-=- who provide a solution to this problem by minimizing what they call inter-group crossings. In their approach, an energy function is minimized to apply a good rotation to their circular drawings to re... |

6 | TopoLayout: Graph Layout by Topological features
- Archambault, Munzner, et al.
- 2005
(Show Context)
Citation Context ...topology. We also describe the HDE approach, so that our HDE detector can be understood. In addition to the work presented here, we have also described some preliminary work on TopoLayout in a poster =-=[2]-=-.sIEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS 2 A. Multi-Level Graph Drawing Algorithms Multi-level methods for graph drawing have been studied to improve algorithm run time with drawings... |

3 |
A meta heuristic for graph drawing. learning the optimal graph-drawing method for clustered graphs
- Niggemann, Stein
- 2000
(Show Context)
Citation Context ...never in a multi-level context. Two previous algorithms search for topology in the graph at a single level, both employing different algorithms depending on the topology detected. Niggemann and Stein =-=[25]-=- describe a multi-level algorithm based on the recursive application of Λmaximization clustering. For each recursively clustered subgraph, the algorithm constructs a feature vector containing statisti... |