Which Steem tags are most commonly used together? Can we learn anything from visualizing the "word cloud" of tags which are frequent neighbors?
I gathered the tags and net rshares for all top-level discussions from the month of May, and started doing some experiments. Here's the 20 tags used most commonly with #steemstem, visualized as a graph with edge widths corresponding to the number of discussions that included both tags.
(Right-click and "view image" to see a bigger version, sorry for not having a better mechanism figured out--- can I make the image a link to itself?)
The numbers in percentages are the proportion of rshares that discussions with those tags received; for example 2.429% of upvote power went to topics under #esteem.
Dialing the neighbor count up to 40 just creates a bigger mess:
Even though GraphViz is told to move tags which are more related closer together, no obvious sub-structure is visible.
Part of the problem may be the presence of non-topic-specific tags like #busy or #esteem. Let's create a blacklist containing #life, #esteem, #steemit, #steem, #busy, and see what happens:
Looks like I should have axed #blog as well.
Now that some of the more popular tags are gone, we can also increase the factor by which edges are widened. It is a logarithmic scale, so using log base 2 instead of log base 10 gives us a substantial difference, which highlights some relationships but doesn't make the graph as a whole any clearer.
This approach isn't totally useless, if we take a low-popularity tag like #math there is some interesting features like "moron" and "dumdum" sitting way off in the corner where they belong.
But there's more work to be done on figuring out how to highlight things that are interesting or surprising. An article (and research paper) I read on this topic is https://medium.com/@uwdata/surprise-maps-showing-the-unexpected-e92b67398865 which I'm still thinking about ways to implement.