Fig 1.
Graph of the 5–communities of the French pre-electoral political twittosphere calculated from August 1st to December 31th 2016.
The multi-polar political landscape is clearly visible with a position on a left-right axis for most political parties. It should be noted that the alliances observed during the political campaign took place between political personalities that were close on this map: Hamon-Jadot, Macron-Bayrou, Le Pen—Dupont-Aignan. The right-wing (Republicans, blue) is divided into several currents which are reflected by different communities. The Juppé community is at the left of the right-wing while the Fillon and Sarkozy communities are close from each other and much more at the right. The latter two merged after the right-wing primary while the Juppé community gradually migrated to the Macron community. The Socialist Party was also made up of a “left-wing” (Hamon community, Montebourg and Pinel community) and a “right-wing” (Holland, Valls). This graph has been spatialized with the Gephi software [36] using the ForceAtlas2 algorithm [37]. See File A1 at DOI: 10.7910/DVN/AOGUIA for the anonymized gexf version of this graph.
Fig 2.
Variation in the number of nodes and links for the graphs of .
Fig 3.
Map of the themes covered in the 2017 presidential programs, drawn-up using Gargantext.
The map has been built from policy measures extracted from the candidates’ programs. The nodes of the map are labels for groups of terms deemed equivalent in political discourse. The link between a label A and a label B indicates that the probability that A and B are jointly mobilized in the same political measure is high. Gargantext applies the Louvain algorithm to identify clusters of labels with strong interaction between them and displays them in the same color. To improve readability, the map was edited in the Gephi software (https://gephi.org) to set the size of nodes and labels according to a monotonous function of their PageRank [50]. File A3 at DOI: 10.7910/DVN/AOGUIA provides an editable version of this map (gexf).
Table 1.
Mapping between political leaders and their party on Feb. 6 2017.
Table 2.
Recall for the followed accounts that were active on the period January 23–February 6 2017 with at least 10 tweets.
Table 3.
Precision for the followed accounts that were active on the period January 23–February 6 2017 with at least 10 tweets.
Table 4.
Recall for the random sample of accounts that were active on the period January 23–February 6 2017 with at least 10 tweets.
Table 5.
Precision the random sample of accounts that were active on the period January 23–February 6 2017 with at least 10 tweets.
Fig 4.
3-communities in the French political environment on the day of the first round of elections (T = [9 April 2017–23 April 2017]).
The environment is highly fragmentary and multi-polar, and the main political forces are represented (the percentages shown correspond to the scores obtained in the first round): Mélenchon and France Insoumise (bright red—19.58%), Hamon and the Parti Socialiste (light red—6.36%), Macron and En Marche! (pink—24.01%), Fillon and Les Républicains (blue—20.01%), Le Pen and the Front National (brown—21.30%), Dupont-Aignan and Debout la France (lilac—4.7%). In the case of the “small” candidates, Asselineau (0.92%) and Arthaud (0.64%) are located inside the same community (purple), Poutou (1.09%) is positioned close to Mélenchon, but does not have a real Twitter community. It can be noted that not only is Bayrou located inside Macron’s community, which is not surprising in view of their unification on February 23, 2017, but also that Juppé, who was previously a candidate for the Les Républicains primary election, is considerably closer to the community of Macron than to that of Fillon and Sarkozy. His “lieutenant” Edouard Philippe, was later to become Macron’s Prime Minister. There are in addition two significant communities that are not labeled with a political leader. Colored orange, to the left of the figure and close to Mélenchon’s community, one can discern a community identified as being associated with the Discorde Insoumise, a video-game community which played an important role in Mélenchon’s campaign. Coloured yellow, at the bottom-right of the figure, close to Le Pen’s community, we identify a community of English-speaking populist and nationalist accounts which strongly supported Le Pen. This community is itself made up from sub-communities, in particular English-speaking supporters of Le Pen, Trump supporters, and UK pro-Brexit supporters. One can also observe significant differences with respect to the pre-presidential campaign of 2016 (Fig 1): in addition to the “left-right” axis, which was dominant at that time, a new (“vertical” on the map) axis appears to be significant, distinguishing between parties with a nationalist, patriotism or protectionist focus and those accepting or adhering to the globalization of economic trade. Note that the fact that the “small” candidates are located at the center of the map should not be interpreted as a form of centrality. Only topological adjacencies are significant on this type of graph. As the “small” candidate communities are not well connected with the other communities, they are located at the center as a consequence of the spatialization algorithm used to make the graph. File A4 at DOI: 10.7910/DVN/AOGUIA provides an editable version of this graph.
Fig 5.
3-communities in the French political environment between the two rounds of elections (T = [April 27 2017–May 07 2017]).
With only two candidates left running for the presidency (Macron and Le Pen), the political patterns of Twitter political activist changed radically with two poles corresponding to supporters of the two remaining candidates (that came from the recomposition of former communities) and a third pole constituted by those who refused to take sides. These latter positioned themselves to be the next opposition force to the future government. It should be noted that the party Les Republicains split during this between-two rounds, the tenors of this party being scattered between Macron and Le Pen informational spheres. File A5 at DOI: 10.7910/DVN/AOGUIA provides an editable version of this graph.
Fig 6.
Representing communities evolution with an alluvial diagram.
Fig 7.
Reconfigurations of political communities between June 2016 and April 2017.
Several major newspaper headlines have been added to shed light on the observed bifurcations. The labels of communities that were still active during the second round have been added in large characters to improve their readability. Each vertical bar corresponds to a political community on Twitter, labeled with the accounts of candidates who were active in the community. The bar’s height is proportional to the number of Twitter accounts belonging to this community. These are computed every Monday.
Fig 8.
Reconfiguration of political communities at the time of the primaries held by the right-wing party.
These primaries were marked by three debates and two votes. One can observe the progressive and regular reinforcement of the Fillon community, week after week, although no-one was predicting his victory in the second round, since Juppé was tipped by the polls to win. As a result of the first round, following the surprise victory of Fillon and the equally surprising elimination of Sarkozy (president of the Les Républicains party and ex-President of France), only a minute portion of the Sarkozy community moved to that of Fillon. The Sarkozy community was reluctant to join the Fillon community, and Sarkozy spent several days negotiating positions for his members within the Fillon structure. The Juppé community, however, immediately distanced itself from the new political orientations adopted by the party. This right-wing split was apparent in the pre-electoral environment (see Fig 1). Following the failed return of Juppé in March 2017 (see Fig 11), part of his community moved to join Macron. The bifurcations observed here are clearly summarized by an article published by BFMTV on 12 December 2016: “Life after the primary: Sarkozy relocates his close allies, Juppé is ‘down’”. Each vertical bar corresponds to a political community on Twitter, labelled with the accounts of candidates who were active in the community. The height of the bar is proportional to the number of Twitter accounts belonging to this community. These are computed every Monday.
Fig 9.
Reconfiguration of political parties when the incumbent president François Hollande abandons the race for reelection.
This declaration was made shortly after Macron had announced his own candidacy. The alluvial graph clearly shows that Hollande’s announcement revitalized the Macron community, at the expense of Valls. Whereas the size of the Macron community had remained stable at the time when he announced his candidacy (427 accounts the previous week, 561 the week of the announcement, 472 the following week), it jumped to 749 accounts during the week when Hollande renounced standing for a second term, and then to 1479 accounts the following week. From that time onwards, this community continued to grow regularly until the elections. This change coincides with a slowdown in the Valls-Hollande community, the right-wing section of the Parti Socialiste that was ideologically closest to Macron. However, the Hamon community, which is further to the left, remained relatively stable during this period. Each vertical bar corresponds to a political community on Twitter, labeled with the accounts of candidates who were active in the community. The height of the bar is proportional to the number of Twitter accounts belonging to this community. These are computed every Monday.
Fig 10.
Reconfiguration of political parties at the time of the primaries of the left-wing party.
Benoît Hamon (36.03% in the first round and 58.69% in the second round) and Manuel Valls (31.48% in the first round and 41.31% in the second round) were the two favorites of the primaries and represented two different currents in the Parti Socialiste, referred to by Valls as the “irreconcilable lefts” during his campaign. Although François Hollande did not officially state his preference for one particular candidate, Valls declared that he had his support (see for example in Le Monde on 4 January, 2017: “According to Valls, he has Hollande’s support”). Indeed, in our reconstruction, during the full period of observation, the Hamon community remains quite separate from the Valls community. For most of the time, the latter community is also the Hollande community, in particular at the time of the primaries, thus revealing the ideological proximity between these two political figures. Following Hamon’s victory, a controversy broke out concerning the commitment made by the primary candidates to align themselves with the winner, whereas some were already clearly indicating their preference for Macron. This indecision is clearly visible on the alluvial graph, with significant movements of activists from the Valls-Hollande community towards the Macron community immediately following the second round of the primary. From the 2,075 accounts making up the Valls-Hollande community shortly before its fusion with the Hamon community on February 13 (Valls also joined the Hamon community), only 959 accounts remained. Significant movements of activists continued to take place between these two communities, until the time of the first round of the presidential election. Although Valls finally gave his official support to Macron on March 29, he was no longer present in the Hamon community after March 13. He started appearing in the Macron community on April 17, and remained until the second round, thus indicating his active support. One can also note in this zoom the rallying of the ecologist Jadot to Hamon on February 23, 2017, which is immediately revealed by the fusion of his community with that of Hamon. Each vertical bar corresponds to a political community on Twitter, labeled with the accounts of candidates who were active in the community. The height of the bar is proportional to the number of Twitter accounts belonging to this community. These are computed every Monday.
Fig 11.
Reconfiguration of political communities at the time of Penelopegate.
Penelopegate arrived at a time when Fillon’s following was weakening as a consequence of a controversy surrounding the investiture for parliamentary elections, which was badly received by Sarkozy supporters. Some of these individuals had already started to move away from the Les Républicains community, as early as January 23, 2017, and had formed a separate community throughout the full duration of this controversy. Various revelations made by the Canard Enchaîné led to major movements of activists between different communities, with the communities supporting Juppé (a potential alternative, i.e. a “plan B”, for Fillon’s candidacy) and Sarkozy being progressively reinforced. On March 1st, the day when Fillon was summoned in the context of his investigation, a large proportion of Sarkozy supporters were in support of Juppé’s return, in a final attempt to push Fillon to resign, which was revealed by an unusual flow of accounts from the Sarkozy community towards that of Juppé. The public gathering that took place on March 5th at the Trocadéro Square (Paris) brought an end to the perspectives of Plan B. Although the Fillon community gained a large number of new activists following this takeover, some of the Sarkozy and Juppé followers kept their distance, and finally joined, respectively, Dupont-Aignan and Macron. Sarkozy himself withdrew for several weeks, which led to his absence from most of the Fillon communities between January 16th and April 10th. It was only from April 7th, that he actively gave his renewed support to Fillon’s candidacy, as was for example mentioned in the Parisien headline on that day “Presidential elections: Sarkozy (finally) supports Fillon”, and could be seen by his constant presence within the Fillon community starting on April 10th 2017. As can be seen, the bifurcations observed in the Politoscope take place at the same time, or in some cases even anticipate comments in the press concerning the activity of political communities. Each vertical bar corresponds to a political community on Twitter, labelled with the accounts of candidates who were active in the community. The bar’s height is proportional to the number of Twitter accounts belonging to this community. These are computed every Monday.
Fig 12.
Reconfiguration of political communities at the time of the televised “Le Grand Débat” on April 4th, 2017.
Philippe Poutou created a buzz during this confrontation when he openly condemned the trickery and corruption of candidates Fillon and Le Pen, by criticizing their anti-system attitudes: “When we are summoned by the police, we have no working-class immunity, we go to the police station.” His remarks were relayed by most of the media, and received more than 1 million online consultations on YouTube. Following this intervention his Twitter community, which until that time had been almost inexistent, and often overlapped with that of Mélenchon, developed considerably (to more than 13,000 accounts) over a short period of time.
Table 6.
Detailed statistics for communities of the 6 candidates with the most intensive use of Twitter in their campaign (Tk ∈ [Feb. 1st 2017–April 22 2017]): Average time spent per account inside the community (< days >), total number of tweets from that community (|Tweets|), number of accounts present at least once in the community (|accounts|) and number of accounts present at least once in the community weighted by the commitment (|weighted accounts|).
Outliers are bolded for greater readability.
Table 7.
Detailed statistics for communities of the 6 candidates with the most intensive use of Twitter in their campaign (Tk ∈ [Feb. 1st 2017–April 22 2017]): Average commitment of accounts in the community (< commitment >); quantities weighted by the commitment: Average number of tweets per account (< activity >), political integration (< integration >) and politicization (< politicization >).
Outliers are bolded for greater readability.
Table 8.
Entropy of the themes addressed by each candidate, and its 3–community.
The two dominant themes as well as their associated percentages are also shown. The maximum entropy is 3.46, if the 11 themes are discussed in a uniform manner. With respect to the selected themes, Benoît Hamon is the candidate who uniformly addressed the greatest number of themes, whereas François Asselineau is the candidate who was the most focused on a subset of themes (including foreign policy and democracy). The variations in attention paid by the candidates to these themes are plotted in Figs I to N in S1 File.
Table 9.
Entropy of the keywords used by each candidate and its 3-community.
The 10 dominant keywords as well as their percentage are also shown. The maximum entropy is 10.64, when the 1596 keywords are discussed uniformly.
Table 10.
Entropy and average number of communities [in brackets] reached on average by fake news, debunks and tweets with or without URL mention over the period.
We also present the result when considering all communities as a single one except for the 6 main political online communities identified during the French presidential election. Tweets without communities (43.44%) were considered as a single community for the processing of entropy. For comparison with the debunks and fakes news dataset, identified by the URLs they spread, we also provide the global statistics for the subset of tweets that mention URLs.
Table 11.
Number of users, number of tweets mentioning a fake news link and the proportion of early spreaders for the tweets containing fake news links.
Early spreaders are users that have been among the first 1/30 percentile of users to share a fake news.
Table 12.
Number of users, number of tweets and the proportion of early spreaders for the tweets containing debunk links.
Early spreaders are users that have been among the first 1/10 percentile of users to share a debunk.
Fig 13.
Radar chart of the normalized propensity to spread fake news (dotted red) and debunks (blue) per community.
1 means a spread at average level.