Click on any node or retweet to display info
Last Retweets
is a real-time network analysis tool that focuses on spanish and catalan politics in Twitter. It enables people to observe the unfolding of the current political trends not only at an individual level - e.g. influence of a node - but also at a systemic level - e.g. emergence of political communities - real-time.

The heart of is the network graph that shows the aggregation of millions of retweets that are filtered with a set of political semantic fields --- each one of them consisting of several keywords. The data can be visualised on two levels: topics and political clusters/communities. Each topic derives from retweets through the semantic fields that are related to a specific political party - e.g. a tweet containing the word 'Rajoy' and 'Aznar' would match the topic 'PP'. Clusters represent the political communities that emerge from the network structure. For example, a sustained decrease in the relative size of a cluster-community may indicate that the political position is losing supporters. The clusters are not based on the popularity of the topics but on the structural properties of the network itself. provides information at an individual level as well. Several metrics such as betweeness-centrality, in-degree, or out-degree among others show the actual influence of a particular individual - e.g. the in-degree metric. The combination of metrics both at an individual level but also at a systemic offers opportunities for further analysis. For instance, a user's specific political sympathies can be inferred by pinpointing their spatial position in the network.

For more information visit the help section.

is based on Node.js, and VivagraphJS


Menu options


Loads a different time-interval subset of the network from the server. Options are last 24h., last week, last month and last year. You can also start a network from scrath by clicking the 'From now on'. A specific interval is not supported at this version but planned for a future version. Note: the time intervals are approximate and usually more recent due to the limited amount of nodes and edges displayed.

Node size

Chooses the metric to be used in order to display the size of the nodes. The In-Degree - number of times a node has been retweeted - or Betweeness-Centrality options are recommended to display the influence of a specfic node in the network. Out-Degree - number of retweets - shows who spreads the information. Best In-Degree / Out-Degree displays in-degree or out-degree depending on which of the two is greater. Followers, Friends and Favourites display the number of people following a node, the number of people a node follows and the number of times a node has been favourited respectively.

Node color

Chooses the network analysis mode. Node colours represent either conversation topics or communities. Topics (political parties) are displayed either as the latest or average topics a node has been talking about. The colour of the node depends on which topic dominates - e.g. a user (re)tweeting 12% of the time about PP and 5% about PSOE will be categorised into 'PSOE'. However, if Average Topics is selected the colour of the node depends on how strong the trend is. Political clusters show the emerging political communities in the network. It also reveals the political affinity of a specfic node depending on its spatial position in the network. The clusters do not always represent a specific political party. For example, the core cluster - Left / Far left - also includes central views and represents the political opposition to the current government. Other clusters such as UPyD are more homogeneous and represent a specific political party. The Catalanism cluster is also an example of a an extremely heterogeneous community that does not represent a specific political tendency - left or right - but a regional political sentiment.


Downloads the current network graph in JSON or GEFX . GEFX is recommended since the network can easily be analysed and visualised with Gephi.

Delete Outliers

If unconnected nodes and/or edges appear - especially with the Load from now on option - they can be deleted from the network using this option. The DFS (Depth First Search) is an expensive algorithm and it is not executed in the background. While the algorithm is executed the application freezes. The processing time might take from a few seconds to a few minutes depending on the size of the network.

GUI and controls

Network graph

The network graph shows the political network downloaded and updated real-time with live data coming from server. Use the mouse wheel to zoom-in / zoom-out. Press R to reset the view. Click (outside of a node) and drag mouse to pan inside the graph. Clicking on any node displays its connectivity and associated information at the user info panel (see below). Blue coloured edges represent in-relationships - retweeted by - and red edges show out-relationships - has retweeted. A hovering selection mode is also available and can be activated by pressing ALT while hovering the network graph.

Search panel

Allows searches by the user's screen_name or name on Twitter within the current network. Search is case and dyachritics insensitive. Server-side searches will be supported in later versions.

User info panel (top right)

The user info panel displays information on the selected node. It shows the user's political sympathies, the list of topics the user has been talking about and their connectivity. Clicking on the picture opens the user's Twitter profile in a new tab/window.

Retweet stream panel (bottom right)

Shows the last retweets processed by the server. Note that in contrast with a regular Twitter stream only the source of the retweets is displayed. The same retweet may appear several times if it is often retweeted. Clicking on the retweets selects the user in the network graph and shows theirs associated information in the user info panel. A fast hovering mode is available. Hovering the retweet stream while ALT is pressed will select the node and display the info without a mouse click.


The legend is composed by two main items. The statistics text is displayed on the top of the legend and gives real-time information about the state of the network - number of nodes and edges on both server and client sides -, the date of the oldest edge visualised and the volume of the processed retweets (per minute). The appropriate legend is displayed under the statistics and displays the categories depending on which of the Node Color options is currently active. The volumes (percentages) shown next to each category represent the amount of nodes belonging to the category. Average topics and latest topics show the percentages that are relative to the displayed network, whilst political clusters show percentages relative to the whole network of the nodes that belong to any of the clusters - i.e. the neutral category does not show any percentage as these nodes do not belong to a cluster.

Final notes


The size of the batches sent by the server has been adjusted to minimise the loading times and keep the essential structural properties of the subset network. With a standard broadband connection the loading time of a standard batch - 1,000 nodes and 12,000 edges aprox. - should take less than 30 s. The visualisation and real-time layout of the network graph should run smoothly in a Apple Mac Book Pro 13' (2009). However, since the network is constantly updated, it is expected that the performance will drop as the network grows. Depending on the hardware specs of the client machine and the activity of the network this might happen in a short period of time or might take several hours before being noticeable. Our tests show that the system is able to render up to 20,000 edges and 3,0000 nodes.

Network batches

Except for the 24h batch which is re-computed every hour, the rest of the batches are re-computed once a day. The metrics such as betweeness-centrality and clusters are also recomputed once a day prior to the batches. The 24h. batch contains half of the edges of a standard batch - 6,000 instead of 12,000. This happens because the core network is prioritised when the batches are computed. In a period of 24h. the resulting network is extremely sparse - lots of weakly connected nodes -; as a consequence the resulting subset has a larger number of nodes and a smaller number of edges than the typical subsets.

Click here to send us any feedback, questions or issues you found at or feel free to write your comments in Twitter

Click on the logo (top-left icon) to open the menu after loading
We are really sorry but your browser lacks full WebGL support. At this time is only supported by Chrome (OSX and Windows), Opera 18 (OSX and Windows) and Mozilla Firefox (OSX only) browsers.
We are experiencing technical problems. We will back soon.