[VIS20] Responsive Matrix Cells: A Focus+Context Approach for Multivariate Graphs

We present Responsive Matrix Cells, a focus and context approach for exploring and editing multivariate graphs. In order to facilitate such analyses, we propose to apply local focus+context regions that embed visualizations within the selected regions of interest. These Responsive Matrix Cells are not static, but can be adapted regarding the level of detail and the representation type. With Responsive Matrix Cells, we can go seamlessly from overview to seeing details on demand to editing data by embedding situation-dependent visual representations directly into a matrix visualization. The basis for our approach is a customized matrix visualization that shows adjacency information in the lower triangular half and pairwise node attribute similarity in the upper half.

00:47 - The Responsive Matrix Cells, are showing either aspects of the nodes or edges, depending in which part of the matrix they were created. Then, visualizations can either be placed in unit cells, where one visualization per cell is shown, or in meta cells, where one visualization is provided for the whole selected region. These two options also define how many data points have to be encoded: one object, two objects, or n objects. Each Responsive Matrix Cell can be freely scaled up, which is synonymous with adjusting the level of detail. Ideally, details are conveyed already at sizes of a few pixels, and when additional space becomes available, it is used efficiently for adding more and more details.

01:30 - In our work, we present specific considerations for visualizations that are suitable to encode multivariate attributes of nodes and edges as well as representations for structural aspects. As an example, we are analyzing a soccer data set from the champions league season ²⁰¹⁷⁄₂₀₁₈. The graph comprises the players as nodes and their connections as edges. The similarity part shows how similar two players are across selected attributes. For the adjacency, an edge exists between two players if they have played for the same club before.

02:04 - To investigate specific players in more detail, we can select the region of interest and create a Responsive Matrix Cell. Here, we decide to start with a unit cell setup. For each cell in the region of the interest, the Responsive Matrix Cells show embedded bar charts, so that the individual attributes and values of the players become visible. We can then scale our Responsive Matrix Cells further. As the size increases, more details become visible in the embedded visualizations.

02:33 - Depending on the analysis task, we can further adapt the Responsive Matrix Cell to the needs of the user. For example, we can switch from the unit cells, which show a representation for each individual cell, to a meta cell that provides a singular representation of our selected region. After looking at the multivariate aspects, we are also interested in the structural properties of the graph. Therefore, we switch to the other half of the matrix and continue our analysis of the edges and their associated attributes. Furthermore, as we prefer the more intuitive node-link representation for reviewing the structure, we switch the embedded visualization to it for this task.

03:18 - An advantage of our Responsive Matrix Cell approach is that if we find erroneous data during the exploration, we can seamlessly transition to editing the data and correct the errors directly in the visualization. During the editing process, not only the current visualization is updated on the fly, but also the other unit cells in the Responsive Matrix Cells as well as the matrix cells in the similarity part encoding the changed data point. Notably, these editing mechanisms can also allow for conducting what- if scenarios. After finishing the editing, the analysis session can be seamlessly continued by discarding and defining new regions of interest with Responsive Matrix Cells. In summary, our Responsive Matrix Cells approach offers a great degree of flexibility in terms of what, where, and how information can be shown in the matrix.

04:12 - For example, unit cells can detail the aggregated value of a single cell or allow comparison in a small multiple arrangement. Meta cells facilitate the exploration of a group of cells, both for multivariate attributes and structural aspects. With this flexibility, Responsive Matrix Cells form a versatile tool supporting analysis workflows going from overview to details to editing. All without the need of switching to external tools or views. .