(Español) @metroaverias: Los ciudadanos como sensores de los servicios públicos


Sorry, this entry is only available in Español.

Anatomy of a Trending Topic

No Comments »

Sorry, this entry is only available in Español.

#MuseumWeek 2014 in depth

No Comments »



#MuseumWeek Support byTwitter arts


From March 24-30, museums across Europe took part in the first ever #MuseumWeek. You can follow all Participating museums across UK, France, Spain and Italy on these discover pages, and join the conversation by tweeting #MuseumWeek

We have done an analysis of the first #MuseumWeek from three approaches, the first from the metrics provided by the t-hoarder tool, the second using network analysis with Gephi tool and the third showing a space-time spread made with a map cartodb

We can conclude that: Read the rest of this entry

The spread of a good tweet


Interesting case study on how a good tweet spreads even if the network is small. The tweet became viral and some media such as The Huffingtonpost, La Vanguardia or La Voz de Galicia published an article about it.

It happened on March 13 when I saw this pretty good tweet on my timeline thanks to a @rafamerino’s RT. I retweeted it too when It was retweeted more than a thousand times. I was surprised to see that the author of the tweet had 92 followers (now, after the success of the tweet, that number is almost double). At that point I decided to study the spread of this tweet :

Note: The poetic report about the stock exchange on the teletext is the work of an anonymous author who signed with the pseudonym Raimundo Diaz and every day publishes an article in Expansión (a Spanish economical newspaper) whit a single word headline. I never thought reading Market information could be so delicious.

Examining in each minute the number of users who spread the tweet and the sum of their followers can presume what users helped to propagate the message. As shown in the figure below, when users with a high number of followers retweeted it, in the next minutes many users retweeted too. However, it is clear that popularity is not the same as influence, since the number of followers was not proportional to the number of induced propagation.

Presumed propagation inducers are:

  1. reaction at 1:00: @lwtuaznar (12,2K followers), @ciudadfutura (13,7K followers) y @javigomezsexta (21,4K followers)
  2. reaction at 8:00 (la mayor): @JotDownSpain (109K followers)
  3. reaction at 11:00: @HiginiaRoig (67,3K followers)
  4. reaction at 17:00: @euribor_com_es (87,6K followers)
  5. rreaction at 22:00: @cdelamorTVE (52,2K followers), @JavierCapitan (102K followers) y @pacoleonbarrios (1.090K followers)

I wonder:

  • Did the timing of the retweet determine the spread?
  • Did the type of popular audience favour the virality of the tweet?


  • We captured tweets containing the phrase “But who writes the rtve Teletext ” with the method search of Twitter API
  • Computing with an own script (in Python) the number of users per minute that retweeted and their followers
  • Obtain from authors’ location declared on their profile the geographical coordinates with google maps API
  • Use Tableau Public to create the graph
  • Use cartoDB with torque to generate the map

Viral on Twitter: TweetMatrioska

No Comments »

This is how a curious chain of tweets encapsulated within each other was spread in space and time. I have called this experiment TweetMatrioska because as traditional Russian dolls, a tweet contained a tweet which inside contained another tweet and so on until the tweet source.

(The map is made with CartoBD, the wizard used was Torque. A true wonder so powerful and easy to use)

It all began on February 14 when I found this chain by @estebanmoro and @moebio. Seeking the source I found the origin, @BenHowe, not without examining nearly 40 tweets before.

This experiment aroused my curiosity because it is closely related to my research on propagation and I quickly got to attempt recovery of tweets from this chain. I found this solution but I wanted to analyze this dynamic diffusion with my own resources and these are the first results of the experiment.

Propagation network topology

  • Number of nodes: 10,568
  • Number of connections between nodes (multiple connections between nodes are counted as one): 11,170
  • Network diameter : 70 ( the largest chain nested tweets was 70! )
  • Average chain length : 20,54
  • Clustering coefficient : 0.006

Temporal propagation
The following timeline represents the number of tweets/hour ( GMT) during the propagation.

To get the data I have followed this methodology:

  1. Find the tweet source from my chain ( 38 links)
  2. From the tweet source obtain by search API what tweets included the URL into the tweet and by a recursive manner get the chain of tweets. I found 11,339 tweets from 10,568 different users. By this method I did not acquire all tweets (I did not get elements from my chain). This could be due to two reasons:
    • The search API does not provide all the messages asked
    • If the URL of the tweet is encapsulated with an URL shortener as bitly or owly, is not found when looking for the tweet as https://twitter.com/screen_name/status/xxxxxxxxxxxxx
  3. Generate a graph with an ad hoc script for this type of propagation
  4. Find the geographic coordinates of the locations declared in the user’s profile with the Google Maps API (has a rate limit of 2,500 per day). The coordinates of the 70 % of users were obtained

  • RSS feed
  • Twitter
  • Linked In
  • Facebook
  • YouTube