Sunday, August 31, 2014

5G on Twitter: NodeXL social network analysis

A NodeXL SNApshot is a great way to catch up with who's saying what about a topic on Twitter. This post discusses the SNApshot that graphs the 1,963 tweets containing the hashtag #5G posted over the period 17 Jul - 29 Aug 2014.

Here's the Gallery Glimpse video:

The video and this blog discusses the page

I used a Twitter search on the nodeXL Graph Server against the hashtag #5G to select the tweets to study. The messages in the graph were tweeted over a 43 day period in late July and August, 2014.

A quick way to see what's been going on is to read the top words in each group label. For example:
  • G1: A 5G network launch in London by 2020 (more about London in G7, G8, G9, ...)
  • G2: News about BellLabs/Alcatel-Lucent
  • G3: More about Bell Labs, and references to Neelie Kroes and the European Commission 
To find the most tweeted stories, which usually explain the words in the group labels, scroll down the SNApshot page to the Top Word Pairs in Tweet lists. The most tweeted URL in the entire graph explains the references to 5G in London: it is a story in the Telegraph about Boris Johnson's pledge that there would be 5G in London by 2020.

It turns out one could pretty much read this off from the list of Top Word Pairs in Tweet in Entire Graph further down the page without even going to the URL:
The Top Word Pairs are less helpful in figuring out the Bell Labs and EU references in the G2 and G3 group labels. The Top URLs in Tweet in G2 provide an explanation: the first URL in the list on reports that a "Bell Labs expert hails future 5G a communications revolution.” The EU mentions turn out to be unrelated (correlation doesn't mean causation!) and are generated by press releases about a speech by Neelie Kroes (Vice-President of the European Commission) about 5G.

Twitter is all about PR, so this graph reveals which flacks have been earning their keep. Scanning the words associated with the groups reveals they work for Boris Johnson, Bell Labs, Ericsson, Alcatel-Lucent and Neelie Kroes, with honorable mentions for the folks from National Instruments, Nokia, FFTélécoms and the CTIA.

To find the most influential twitter users in this area, scroll down the page to the Top 10 Vertices, Ranked by Betweenness Centrality list:
The international character of the topic is shown by the fact that the Top Words on the graph are not only in English:
G10, G13, G15: French
G11: Japanese
G17: Russian
Perhaps the most curious element showed up on the Top Word Pairs in Tweet in G5 list:
What was the coincidence? Looking at the URLs for G5 reveals that "ni" is the measurement company National Instruments (no relation to Monty Python), seller of the LabVIEW software package, and "niweek" is the company's 2014 annual conference. However, the URLs themselves don't point to anything that reveals the coincidence, though there are quite a few National Instruments stories:
To solve the puzzle required digging into the tweet record in the Excel file for the graph (click Download the Graph Data as a NodeXL Workbook at the bottom of the Graph Gallery page; to make best use of the file, one needs to install the NodeXL add-in). The file reveals a tweet originated and promoted by the National Instruments "@labview is used by the top 20 #5G researchers... Coincidence? #NIWeek". So the Graph Gallery doesn't reveal all: sometimes one has to go to the source data.

In terms of structure, this graph is a Community Cluster using the taxonomy of the Pew Research paper Mapping Twitter Topic Networks. It shows the characteristic large group of disconnected contributors who mention the topic but do not link to one another (group G1, with about 30% of the vertices), and loosely interconnected sub-groups that focus on subtopics.

To get a sense of the structure of the graph, easiest to click on the "View an interactive version of this graph (experimental)" link that takes one to an X-ray of the graph

I've color-coded the vertices from blue to red by increasing eigenvector centrality (CITE) and one can see that there's a fair amount of cross-referencing between groups G2, G3, G4 and G6. (Eigenvector centrality highlights users that are well connected to other well-connected users; Activate Networks has a good introduction to centrality metrics.) Interestingly, though, group G5 (the nodes of ni) is disconnected; the National Instruments crowd talked about 5G, but wasn't connected to rest of the conversation.

No comments: