E VOLUTION OF COMMUNITIES ON T WITTER DURING THE 2017 F RENCH PRESIDENTIAL ELECTION No´ e Gaumont, David Chavalarias, Maziyar Panahi Institut des Syst` emes Complexes de Paris IdF & CAMS http://politoscope.org September 2017 N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 1/13
Observation • The 2017 French presidential election was unconventional and surprising. • Twitter is part of the communication strategy of every major candidates. N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 2/13
Data - collected since June 2016 Following ≈ 3700 accounts of politicals figures with the track and follow Twitter API). Analysed Not collected (~50% à 80%) followed account : X @X - #X - 'X' Not followed account : Y T(X,A) le 8 juin 2017 # Tweets : > 60 M # Unique Tweet(Y,B) T(Y,T(X,A)) user: > 2 . 4 M Not followed account : Z T(X,T(Y,B)) T(Z,T(X,T(Y,B)) T(X,A) : The account X tweets content A N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 3/13
Definition of social communities Lots of variables : • The follower/followee relation • The retweet relations • The semantic proximity • The mentions @ • The reply links • ... We choose to first analyse the dynamics of communities. The retweet relations is dynamic and is less ambiguous than mentions and replies. N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 4/13
Definition of social communities Goal : Identification of political communities though the graph of retweets. Louvain + strong link N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 5/13
Communities detection on which time window From August to December From August to February N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 6/13
Communities detection on which time window From August to December From August to February Too short → empty graph Too long → clique-like graph N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 6/13
Retweet time 700000 1 0.1 0.9 0.08 600000 0.8 0.06 0.04 0.7 500000 0.02 Number de retweets 0.6 CDF 400000 10 20 30 40 50 0.5 In days 0.4 300000 0.3 0.2 200000 0.1 0 5 10 15 20 100000 Retweet time in hours 0 Monday Tuesday Wednesday Thursday Friday Saturday Sunday Retweet time converse Weekly pattern of retweets cumulative distribution N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 7/13
Retweet time 700000 1 0.1 0.9 0.08 600000 0.8 0.06 0.04 0.7 500000 0.02 Number de retweets 0.6 CDF 400000 10 20 30 40 50 0.5 In days 0.4 300000 0.3 0.2 200000 0.1 0 5 10 15 20 100000 Retweet time in hours 0 Monday Tuesday Wednesday Thursday Friday Saturday Sunday Retweet time converse Weekly pattern of retweets cumulative distribution Two weeks sliding periods for the retweet graphs. N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 7/13
Community Evolution N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 8/13
Community Evolution N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 8/13
Focus on the right primary Fillon and Jupp´ e lead the first round Fillon won the second run N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 9/13
Focus on the right primary Fillon and Jupp´ e lead the first round Fillon won the second run N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 9/13
Community as feature for other study N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 10/13
Community as feature for other study N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 11/13
Conclusion lien vers le papier N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 12/13
T HANK YOU FOR YOUR ATTENTION No´ e Gaumont noe@ngaumont.fr Works done with : David Chavalarias, Alexandre Delan¨ oe, Jean-Baptiste Leenart, Maziyar Panahi politoscope.org iscpif.fr N O ´ E G AUMONT , I NSTITUT DES S YSTEMES C OMPLEXES DE P ARIS I LE - DE -F RANCE - HTTP :// ISCPIF . FR ; 13/13
Recommend
More recommend