The Network Lens: Interactive Exploration of Multivariate Networks Using Visual Filtering Ilir Jusufi, Yang Dingjie, and Andreas Kerren School of Computer Science, Physics and Mathematics (DFM) Linnaeus University SE-351 95 V¨ axj¨ o, Sweden Email (corresponding author): ilir.jusufi@lnu.se Abstract—Networks are widely used in modeling relational data often comprised of thousands of nodes and edges. This kind of data alone implies a challenge for its visualization as it is hard to avoid clutter of network elements if using traditional node-link diagrams. Moreover, real-life network data sets usually represent objects with a large number of additional attributes that need to be visualized, such as in software engineering, social network analysis, or biochemistry. In this paper, we present a novel approach, called Network Lens, to visualize such attributes in context of the underlying network. Our implementation of the Network Lens is an interactive tool that extends the idea of so-called magic lenses in such a way that users can interactively build and combine various lenses by specifying different attributes and selecting suitable visual representations. Keywords-graph drawing; network analysis; multivariate network visualization; magic lenses; interaction techniques; I. I NTRODUCTION The visualization of complex and large networks is aimed to give insight into different patterns between relations of data. Most of these networks are represented following a node- link metaphor comprised of thousands of nodes and edges. Visual analysis tools have to deal with huge, complex and dynamic data sets and cope with different challenges, such as to avoid clutter and to increase the people’s understanding of graphs, also known as readability of the network [1], [2]. In practice, each network object may additionally have a number of attributes that are important to be visualized in context of the overall network presentation. Finding a good solution to visualize those attributes is an ongoing challenge in various network visualization domains, such as software engineering, social network analysis, or biochemistry. One of the most simple software engineer- ing examples is the visualization of relationships between classes. Each class may have a number of methods, fields, and other properties. In addition, we could compute different measurements (software metrics) for such elements that are important to maintain and to improve the software engineer- ing processes. Examples for popular software metrics are lines of code, number of classes, etc. From the perspective of the visualization community, the measured values of a whole software metric suite form a multivariate data set. To give an additional example from social network analysis research, we could imagine that the relationships between staff members of a large company should be analyzed. The set of all relationships forms a network. Of course, each staff member also has an individual set of own properties, such as age, gender, qualification, position, etc. The question is now: how can we visualize those attributes together with the network drawing? There are a number of approaches to address this problem. The simplest one is to present a list of all attributes and their values in a separate view on the display. This could be a textual list or a more complex visual representation, such as parallel coordinates [3] or star plots [4], as the attributes form a multivariate data set. Another way is to use glyph- based approaches where we can represent attributes by using visual features of the glyphs, e.g., shape, orientation, color, or size [5]. A more detailed presentation of related work is given in Section II. In this paper, we will discuss an extension of the traditional magic lens idea (cf. Section II-C), called the Network Lens, applied to traditional node-link graph layouts. We have developed a prototype implementa- tion of this Network Lens that enables users to interactively build various lenses by specifying different attributes and selecting different visual representations [6]. Each time we apply our Network Lens on a network element, it visualizes its attributes (or a subset of them) by using a specific visual representation, i.e, the standard node representation is replaced by a new visualization or diagram. A neat example would be the one of time-depended attributes, which is a standard problem in biochemical network analysis. A domain expert could analyze experimental data measured over time, which are attached and represented by the net- work nodes at time step t i , for example. Without changing his/her current visualization setting, a specific Network Lens instance could be used to show the data at a time step t i-1 for a set of interesting nodes. In this way, our approach can support the visual analysis process of multivariate networks by having an additional generic tool that can be adapted to standard visualization tools and can extend already existing views to show node and edge attributes of the underlying network. Users are able to create individual lenses, to store 2010 14th International Conference Information Visualisation 1550-6037/10 $26.00 © 2010 IEEE DOI 10.1109/IV.2010.15 35 Information Visualisation 1550-6037/10 $26.00 © 2010 IEEE DOI 10.1109/IV.2010.15 35 2010 14th International Conference Information Visualisation 1550-6037/10 $26.00 © 2010 IEEE DOI 10.1109/IV.2010.15 35 © 2010 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE. DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/IV.2010.15