I gave a presentation at the University of Calgary's Collaboration for Learning conference today, on some of the visualizations I built as part of my thesis research. I made a point of avoiding talking about the thesis itself, but presented some of the key visualizations of metadata and coding data. I also made a point of only having enough slides to last for no more than half of the allotted time, in order to ensure enough awkward silence to hopefully prompt an active discussion. Kind of worked, almost.
The presentation was intended to show what kind of information can be gleaned from examining the system-generated or -inferred metadata (title, date, author, wordcount, etc…), and contrasting that with what can be learned by "cracking open" the posts and conducting a latent semantic analysis using a coding template. The conference theme was "collaboration for learning" - so I was trying to take a slightly different angle, to see if it was possible to show what collaboration might look like by analysing online discussions.
Some of the points I made during the setup:
- normalizing online discussion data across platforms is hard, labour-intensive, and not likely to be done by anyone who isn't a desperate grad student trying to finish a research project before running out of time in their MSc program…
- looking at the metadata can be surprisingly enlightening - especially when mapped in a timeline view. Why on earth don't more online discussion analyses use timeline views rather than coarse aggregations at the week/month/semester level?
- pretty pictures are impressive, but often don't actually tell you anything. I'm looking at you, Wordle.
Some of the points that came up in discussion:
- the coding-data analysis may not be necessary to learn much of what can be inferred through more automatable metadata analysis, especially when combined with sources of data (like, radically, talking to the participants…)
- having better coding-data analysis tools may not be as awesome as it sounds, as there is the potential for having nasty feedback loops if the discussion analysis is available to participants during the discussion itself.