Figure 1.
Results of the elections and Twitter volumes.
On top: the pie-chart of the results obtained in the election by the parties. Then the forecast from the Twitter time series frozen at three different final times. Below: the cumulative number of tweets for the various parties and their daily evolution. The yellow and green vertical lines represent the election day and the day after respectively (when exit-polls are released).
Figure 2.
Relative Support from the Twitter volumes, entire country.
Top left: the plot of the RS parameter computed on an average of 10 days centered on the specific day on the x-axis; top right: the values of the RS parameter with no averaging day after day. The parties have been ordered according to their final rank at the elections, with the exception of the ratio SC-SMALL. Note how most of the values are above 1, correspondingly SC-SMALL is below one. Below we have a detailed plot of the tweet volume in the final days before the elections, where changes of linear slopes are especially evident during voting days (yellow in all the figures). The yellow and green vertical lines represent the election day and the day after respectively (when exit-polls are released).
Table 1.
The second column reports the official data from the Italian Interior Minister; these data, taken from http://elezionistorico.interno.it/index.php?tpel=C, have been attached in the Supplementary Information.
Table 2.
The error is the statistical error on the various values computed.
Figure 3.
Spider plot of TV shows, and topic relevance for each candidate.
In this figure we show two spider plots related to the candidates and two main hashtag categories: TV shows and the main terms used in the political discussion shared by all the candidates. Amongst those we selected the 8 most popular. The number of tweets obtained by each candidate associated with a specific hashtag are reported on a logarithmic scale on the relative axis. Mr. Berlusconi outperforms all the others in the TV show section, while in the other case the concepts and words attached to Mr. Bersani (left-wing party) are not that different from those related to Mr. Monti (moderate party), with the notable exception of the word “lavoro” (employment). In both cases the spider profile of Mr. Grillo is sensibly different from all the others.
Figure 4.
Relative Support from the Twitter volumes, Northern Italy.
For the case of the the macro-area Northern Italy, top left: the plot of the RS parameter computed on an average of 10 days centered on the specific day on the x-axis; top right: the values of the RS parameters with no averaging day after day. The parties have been ordered according to their final rank at the elections at the national scale. Since parties ranking differ locally, some of the plots are actually below 1. Below we have a detailed plot of the tweet volume in the final days before the elections, where changes of linear slopes are especially evident during voting days (yellow in all the figures).
Figure 5.
Relative Support from the Twitter volumes, Central Italy.
For the case of the the macro-area Central Italy, top left: the plot of the RS parameter computed on an average of 10 days centered on the specific day on the x-axis; top right: the values of the RS parameters with no averaging day after day. The parties have been ordered according to their final rank at the elections at the national scale. Since parties' ranking differ locally, some of the plots are actually below 1. Below we have a detailed plot of the tweet volume in the final days before the elections, where changes of linear slopes are especially evident during voting days (yellow in all the figures).
Figure 6.
Relative Support from the Twitter volumes, Southern Italy and Islands.
For the case of the the macro-area Southern Italy and Islands, top left: the plot of the RS parameter computed on an average of 10 days centered on the specific day on the x-axis; top right: the values of the relative supports with no averaging day after day. The parties have been ordered according to their final rank at the elections at the national scale. Since parties' ranking differ locally, some of the plots are actually below 1. Below we have a detailed plot of the tweet volume in the final days before the elections, where changes of linear slopes are especially evident during voting days (yellow in all the figures).
Table 3.
The second column reports the official data from Italian Interior Minister.
Table 4.
The RS is computed on an ensemble of 192 points starting retrospectively at two days before the election day.
Table 5.
In this table we present the cosine similarity between all the possible couples of politicians present in the spider plots.