University of Konstanz, Computer Science
|
Sunbelt XX: Sessions on
Network Visualization |
The 20th International Sunbelt Social Network Conference (Sunbelt XX) is held in Vancouver, Canada, on April 13-16, 2000.
To visualize a large network we usually decompose it into several smaller (up to some hundreds of nodes) manageable parts. If such part is sparse enough we can obtain satisfactory representation of it using some standard graph drawing techniques (spring embedders, eigenvector methods,...). To obtain the global structure of a large network we apply some clustering methods (blockmodeling, cores,...). We get a smaller network that is often quite dense. In this case and in the case of denser parts of network the matrix presentation is preferable. Three different matrix displays of network can be used: complete reordered matrix (using reorderings some patterns can be found), image matrix (only relations among clusters are shown). contextual matrix (in image matrix some clusters are shown in details). Additionally, shadowing or colors can be used to represent the density of lines among clusters. The proposed approaches are supported by program Pajek. In the paper some typical examples will be also given.
Social networks are created and explored in an extremely wide variety of domains, and programs for their visualization and exploration need to be very flexible. We present preliminary experiences with Open Krackplot, a visualization program that allows the user to view and modify the algorithms used for network layout and presentation, allowing users to customize the program in ways that would otherwise require assistance from its programmer. We make use of recent work in knowledge representation and ontologies to structure Open Krackplot so that it is relatively easy to modify, and use an English-paraphrase based editor to shield the user from programming constructs. In the presentation we will show examples of modifying both network presentation and layout in Open Krackplot to support the needs of particular studies.
We demonstrate an interactive software tool for network visualization that is currently developed at the University of Konstanz (with contributions from several other places). In addition to a variety of conventional layout methods, it supports the contextual visualization of centrality and prestige (introduced at last year's Sunbelt). The tool is therefore well suited for the interactive graphical analysis of prominence.
One of the interesting discussions surrounding the analysis of network data has been the choice of measure for a particular construct, and why. For example, in a study of 41 differentially effective elementary schools the leadership of the principal was evaluated using centrality measures. It was theorized that more effective principals were more centrally located within the overall faculty network and that this location would be different for effective and ineffective schools. Results from previous research indicated that there are mean differences in the leadership positions of the principals within differentially effective faculty networks, but these differences were not significant at the p.05 level on all measures of centrality - degree, betweenness and closeness. Within the context of other network parameters, however, sociograms did illustrate differences in the location of the principal within the network and those to whom the principal was connected. Past research has also indicated that the definition of the network may also influence the choice of measure. This study further explores these two avenues, the differences between the measures and the impact of the definition of the network on the results and uses the settings found in the Krackplot software package to illustrate the differences for visualizing the results. This paper will describe the results for the 40 elementary school principals, on each measure of centrality, with the corresponding sociograms.
Blockmodels compute partitions of equivalent actors. Adjacency, similarity or distance matrices as well as the image matrix of the block densities can used and combined to produce images. We use zone symbols, convex hulls and spatial aggregations for alternative designs and compare their informational content and readability.
In a series of studies we are investigating the effects of sociogram design elements on human perception of social network attributes. Sociogram design elements can include the layout of nodes, the depiction of individual nodes and edges between them and other features such as the use of three dimensions and motion. Here we report on a study of the effects of motion on perception. In the study, participants were shown a graph in three different time periods. Half the participants were shown smooth motion between the three graphs and half were shown three snapshots. We report our findings on the conclusions that they drew.
Focus is given to two types of networks: different networks defined on the same set of actors; and networks that are changing over time. Different relations are often measured on the same set of actors; in this way several networks are obtained. (Thurman Office, Bank Wiring Room, ...) Time component is often a part of social networks. There exist examples where connections among actors appear, change in intensity or disappear over time. In some occasions even presence or absence of actors change over time. For example: changes in signed graphs over time (Sampson monastery data), changes in HIV networks, relations among actors in different episodes in movies, ... In this paper we present some approaches to visualize sequence of networks implemented in program Pajek. Some typical examples are also given.
New analytical software has improved the visualization of networks. Trained in the recognition of patterns, our eyes now may play an important role in the analysis of social networks. Visual inspection of a graph seems to be a promising way to explore its structure, especially when substantive theories do not offer detailed structural hypotheses that permit the specification of mathematical models. However, our eyes may fool us, so we must make an effort to validate our visual analysis formally. This paper presents a visual analysis of a social network and scrutinizes its results. The network contains the institutional affiliations of experts that advice on the granting of subsidies to Dutch visual artists. Inspecting its structure over time and taking into account the institutions' characteristics, the network is judged to contain three parts in 1997: a tightly knit center divided into a circuit of talent scouts and a circuit of discourse specialists, and a loosely connected periphery dominated by regional clusters. To the eye, the structure seems to be characterized by a 'structural hole' between discourse circuit and regional periphery. In order to evaluate the visual results, the network's hypothesized structure is tested by means of block modeling.
In this presentation, we demonstrate the same data set analyzed with two different methods: social network analysis (SNA) and neural network based analysis. The aim was to compare benefits of different methods in analyzing complex structures of communication networks. Our study concerns a log file recorded in a computer supported learning environment shared by 21 participants. The data consists of over 6000 actions like commenting other participants texts and linking document to another document. With SNA we have calculated the data with three centrality measures. We have also visualized the distances between the participants in a 3D VRML model using MDS. In the neural network based analysis we used Self-Organizing Map (SOM) for visualizing document and user profiles. With SNA it is possible to make a general overview to this communication network and we could also find the most active actors or group of actors in the environment. The information about who makes comments about whom shows the discussion areas. With SOM we clustered similar documents according to action types. Furthermore, we used SOM for revealing clusters of similar actors in terms of their document profiles. It can be concluded that these different methods are able to reveal various levels of information about social networks.