We looked inside some of the tweets by @DWShaffer and here's what we found interesting.
Inside 100 Tweets
Check out our new cool site: http://www.tumgir.com
1 month to submit to the inaugural International Conference on Quantitative Ethnography — join the network coming together around deriving culturally contextualised meaning from large human activity datasets http://icqe19.org #learninganalytics #learningsciences
Not quite a month now, but still time. It's a great chance to people to learn more about Quantitative Ethnography and/or get feedback on your work. Workshops included with conference registration, and there is a doctoral consortium (funded!) as well as an early career workshop.
Hey! Look at the latest from the eminent @sbuckshum: https://learning-analytics.info/journals/index.php/JLA/article/view/6409/7153 …
Even more to the point, people working with other models (for whatever reason) can visualize them without creating an arbitrary layout of nodes with link weights indicated by numbers *and* without introducing arbitrary visual artifacts that confound interpretation of results./23
I think it is an open question, still, as to whether order in this sense matters more than co-temporality. But now we have a good tool to assess that in a direct comparison, rather than comparing two different techniques, with the obvious confounds that introduces. /21
So state transition diagrams, darling of NLP folks, can easily be represented and visualized in a way that lets us quickly model and compare larger number of diagrams. Instead of looking at just the average diagram for a group, we can compare groups, means, individuals, etc./20
So you can see the problem: If we have separate links A -> B and B -> A, what do we do with the co-registration? We've tried two things in the past. First, just treat A -> and -> A as separate nodes. That works mathematically, but is hard to interpret. /15
The co-registration works because ENA uses an optimization algorithm (developed by my brilliant friend @JeffLinderoth) that positions the nodes of the network graph so that the centroid of the network graph approximates the position of the point in the network space. /14
The key to making this work is that network graphs are co-registered with the space, meaning that changes in the graph (more connections between nodes in the "upper" part of the graph) correlate with changes in position the network space (larger values for the y-coordinate)/12