Learning Analytics for Assessment: Ambient displays and exploring datasets

Ambient displays are artefacts used to arise awareness about certain data not directly observable in the physical world (e.g., learning progress), and integrated in the environment. That is, they are not the main focus of the users activity but display information that could be consulted when needed, that make them aware of something interesting for their activity. In the concrete case of ambient displays for learning, several use cases have been analysed in the awesome Dirk and colleagues' literature review. Some conclusions could be drawn out of their work, for instance, that ambient displays if well used have been demonstrated not to distract the students. And that generally the amount of data displayed is quite low. The most sophisticated displays got to make what they called situational awareness, that is, besides acquiring some knowledge and understanding of the situation, the users could obtain a pinch about the future; that is, the data displayed if correctly interpreted could predict future events.

Therefore, I'm playing around with some datasets to come up with a suitable visualization for an ambient display: my dataset from the classon use case; another dataset from tinyarm, a tool for awareness of researchers; and, in the near future, with a dataset from stepUP. I have to first explore the dataset in order to determine how the visualizations should look like (bottom up approach). But I'm also exploring some tools for creating data visualizations, in order to get to know some novel visualization approaches that could be applied to the data (top down approach). Let's see what could I find in the mid-way, supported by the tools and the data itself.

In this regard I wanted to quickly visualize the tinyarm data, provided to me in JSON. I could not find a suitable tool, so my best chance was to export it into CSV using a JS script (I'm using JS for absolutely any processing I need, executed with node). I was pointed to a couple of tools to explore: on the one hand gephi, but it is more about networks (nodes and relations), not so useful for visualizing time series; and gapminder, that could be used to nicely visualize data about countries in the world, but you cannot upload your own data. Anyway, they point to another solution in the FAQ: to use Google Public Data Explorer, where you could upload your data in Dataset Publishing Language. For using this format, you should have your data in CSV plus a XML file describing your data: concepts (organized in topics) and their relations (slices). I can say that constructing the XML is not so easy (and so funny). Besides, I realized that there's a simpler and more powerful tool that I could use: Google Spreadsheets. So, after processing the data with the JS script I'm uploading it into a spreadsheet and visualizing the data using the built-in charts, very powerful to visualize this dataset (around 1.5k rows).

Regarding the visualization of the datasets for the ambient displays, my co-advisor @rcg0 pointed my to a very interesting blog post about visualization of speedos and tachometers. Roughly, it describes why the "in principle not convenient" circular visualization of the speedo make a lot of sense to detect acceleration. At the end of the post a visualization of the last BMW dashboard is presented, and the circle of the tachometer and speedo are linked by a fuzzy line to indicate that both measures are related, and this relation is important in sport mode (in this mode the link appears). That relation could be used as a metaphor for the relation between effort and outcomes (quality outcomes), since a great effort should imply a higher speed if controlling the gears correctly. Sometimes the students claim to devote a lot of effort in a course and the outcomes are not the ones they expect. By means of this dashboard they (and we as teachers) could detect easily these situations and we'll be able to help them to change to the correct gear.