Following successful use of the first museum collection visualization exercise in an actual course, we worked again with collaborator June Abbas in the School of Library and Information Science (L&IS) to develop two further versions. For both our main goal continued to be to help students learn about the basics of visualization, the new interactive data editing features being developed in this project, and the benefits of both for the organization of information and knowledge resources in their discipline.

We set out to extend the features and improve the usability of both the visualization and the exercise itself. We also aimed to better demonstrate the data editing capabilities being developed in this project by more explicitly integrating data editing features as well as observation and interpretation opportunities into the exploration and analysis task flows that users are likely to follow in the visualization in typical usage situations, in this case the curation and exhibition of physical artifacts in museum collections.

We developed version 2 for use in the Fall 2019 offering of the LIS 5043 course. Its development entailed a variety of major additions and changes to both the visualization and the exercise task set. We also created a (silent) video tour of the visualization to demonstrate the various components and interaction describes in the instructions. All students in the Fall 2019 course completed version 2.

We then developed version 3 (shown below) for use in the Spring 2020 offering. Changes in version 3 were limited to minor adjustments to the visualization to better accommodate screen size limitations, plus corresponding modification of the exercise instructions. To support diverse learning preferences and allow flexibility of effort during the COVIC-19 pandemic, version 3 was offered as one of several alternative assignments in the Data Stewardship and Visualization module of the course. Roughly a quarter of the class chose to do the museum collection exercise for the assignment.

The major design improvements and refinements in version 2 of the visualization follow.

• The barchart views in version 1 were little used. We replaced them with three new views. For a selected categorical dimension, a timeline view at the top displays small multiple timelines for each unique attribute value, with the time span of each collection item shown as an oval. For two selected categorical dimensions, a heatmap view in the middle displays a 2-D matrix with a grid cell per pair of dimension values, using a yellow-to-red color scheme to show collection item counts in each cell. A parallels view at the bottom displays a parallel coordinate plot of four selected quantitative dimensions, with one line per collection item.
• For interaction, we implemented two modes of selection (brushing) in addition to the pan-and-zoom modes of navigation typical for each view type. Multiple selected items in the timeline and parallels views are shown in blue, and also highlighted in blue in the central table view. A single selected item in the table is shown in red, and also highlighted in the timeline and parallels view, allowing identification of the selected items relative to all others both in time and in physical dimensions.
• In response to many requests for commenting and tagging features, we added a Notes field and three Tag (A, B, C) fields to the editor interface (top left) and to the summary of object details (bottom center). We added four purely user-editable attributes to the data in the form of four new columns without pre-existing values.
• In response to uncertainty about using the import/export/revert options, we moved the button box to the top left corner to be more prominent, and provided a clearer up-front explanation of their use in the instructions.
• In response to confusion over the meaning of the "Push" and "Pull" buttons in the editor, we replaced them with "Access" and "Modify". These terms are more familiar to L&IS students.
• To the instructions we added descriptions of the timeline, heatmap, and parallels views and how to use the brushing interactions to highlight items in them. We also refined the instructions for using the data editing interface and expanded the descriptions of the various other components in the visualization layout.

Many students inquired about the possibility of exploring larger data sets. Having a larger data set was also needed to make good use of the new visualizations; identifying and characterizing patterns in very small data sets can be difficult. This problem is especially pronounced in the heatmap view when small data set sizes result in nearly all cells being empty or singletons. For versions 2 and 3, we selected and added metadata and photographs of 44 new items to bring the total size of the collection data set to 65.

Anticipating that some students would not have access to large screens, we carefully designed the views and their layout to be as compact as reasonably possible. Doing that entailed designing the timeline and heatmap views to overlay labels of the chosen categorical attribute values directly on top of the underlying time range ovals and cell fill colors, respectively, rather than placing the row and column labels off to the sides. Even with careful choices of text size, font, and color for the labels, reading them can be difficult when a chosen categorical attribute has more than eight or so different values in the data.

Despite our attempts to make the visualization compact, a few students were unable to see the entire visualization on their available devices during the Spring 2020 exercise. For version 3 we added a last-minute optional scrolling feature to the Improvise window to allow scrolling over the entire visualization layout. (Being implemented in Java Swing, Improvise windows can be resized but their contents cannot be scaled to zoom in and out. This shortcoming is one of the motivations for beginning development of JavaFX version of Improvise and its views, since JavaFX window contents can be readily scaled.)

We updated the assignment exercise in parallel with the various changes to the data set and visualization design listed above. We revised several of the exercise tasks to require interaction with more complex data relationships. We focused much more heavily on learning and interpreting the three new views. To do that, we adjusted several tasks to either require the use of the three new views to perform them, or make it much easier to perform them correctly and/or efficiently by using the views. We also added two new tasks that directly entail entering values of tag attributes to express observations about relationships between collection items, for instance to tag a preferred subset of items with a common label based on some user-defined grouping scheme.

Try it out! The Museum Collection Demo is an Improvise visualization. Improvise itself runs as a self-contained, cross-platform Java desktop application. Download the visualization here. Note that your system must have a recent version of Java installed (1.8 or newer) for the application to work. You can also download just the PDF of the instructions/tutorial/tasks here. The download also contains the (silent) video tour shown below.