Sunday, December 28, 2014

Six weeks of spectrum auction tweets

I created an animated GIF to show how twitter traffic about spectrum auctions changed over the first six weeks of the AWS-3 auction, i.e. November 15 to December 27.

The graphs are generated through an analysis in NodeXL of the results of a search for tweets that contained the terms "(spectrum auction) OR (incentive auction) OR (airwave auction)". There were a total of 7,403 tweets by 4,094 users.

To drill down on the graph detail for each frame, here are the SNApshots from the NodeXL Graph Gallery, with characteristic word pairs/strings (see "Top Word Pairs" in the SNApshots; my capitalization) in brackets:

Whole period (34 billion, record bids)
Week 1 (commencing February 23, inter ministerial group forms, 30 billion)
Week 2 (record 34 billion, record bids)
Week 3 (3G February TRAI, defence requirement CDMA)
Week 4 (surpasses 41, FCC AWS3)
Week 5 (Canada unveils plan, 44 billion)
Week 6 (give government unexpected gift 44 billion)

While most of the traffic is evidently about the AWS-3 auction in the U.S., the search has also picked up some other stories, such as the planned February auction of 3G spectrum in India.

As time passes one can see various users come and go as they tweet about this topic. However, a core group keeps chugging through the whole period. (The animation was made by locking the vertices in their positions generated by NodeXL for the full period, and then subsetting vertices that tweeted -- i.e. had edges -- in the successive weeks.)

The most connected users over the entire period, measured by betweenness centrality, are a mix of news services (@wsj, @bw, @theeconomist, @fiercewireless), users in India (@helal_786, @ratanmaitra, @yr_conscience, @dramoly) and the usual US spectrum suspects (@fcc, @mark_goldberg, @tmfassociates, @sutherla, @waltbtig).

Since betweenness includes nodes that are mentioned by many others but don't tweet often themselves (e.g. @fcc), a "reach" metric such as number of tweets ("out-degree", in the parlance) multiplied by the log of a user's followers ("log" to take into account the diminishing returns of many followers); by this measure, the top ten influencers are:


There is a reassuring overlap with the betweenness stars, but spectrum commentators are more heavily represented.

Filtering the graph to just show users that converse with each other yields a much smaller group: just 64 users. Most of them are in pairwise dialogs, but a few them (including some of the Usual Suspects) chat in small groups. (The thumbnails are scaled to the number of tweets.)

No comments: