Date(s) - 17/09/2017 - 22/09/2017
Categories No Categories
Evolution of communities in twitter during the French presidential election
Noe Gaumont, Maziyard Panahi, David Chavalarias
Complex Systems Conference 2017
Site web : http://ccs17.unam.mx
Twitter acquired a central place as a mean of communication for political parties in the last year. For example, Donald Trump has 16.1 million follower on twitter while on a smaller scale François Hollande has 1.9 million follower on the 17th march 2017. This allow them to reach a very large audience. Not only is Twitter a large scale broadcasting platform, it also allows people to react easily with replies, mentions, quotes and retweets. Thus, the impact of each tweet can be measured by the analysis of all the reactions it generated. In this study, we are focusing on the French presidential election and the evolution of the supporters of each candidates during the campaign. Through the Twitter streaming API, we are collecting all the activity surrounding a list of 3 500 accounts, from august 2016. These accounts correspond to deputies, mayors, senators and all the candidates in the French presidential election. This collect includes all the tweets from these 3 500 accounts but also the reactions1 (replies, quotes and retweets) generated by these tweets. Thus on 1st of March 2017, we already collected more than 21 million tweets from more than 1.2 million unique user. Parts of the data are available: multivac.iscpif.fr. Most of the studies on Twitter uses the information of followers/followees to deduce political support, this is a powerful approach and led to meaningful insights on the structure of the Twitter network such as the existence of community structure. However, follow information is binary and not very dynamic. We use the notion of retweet, instead. A single retweet may not be a sign of agreement, however multiple retweets in a short time period are evidences of a strong relations between two persons. Like previous study, we find evidence of community structure inside the
weighted graph of retweets because people close to a candidate hardly ever retweet people from other communities. As retweets are highly dynamic, we are able to have a more fined-grained description of the structure by analyzing the temporal network as a series of graphs on overlapping time windows. By applying community detection algorithm on each graph, we follow how the communities grow, split, merge or decline over time, see Figure 1. The novel ty of our approach is being able to track these evolutions to existing events. This reflects how much the twitter medium reacts to event in the real life (official annoucement, debate, presidential primary). For example, Figure 1 focus presidential primary of the right involving mainly Fillon, Juppé and Sarkozy. After the first round won by Fillon and Juppé, Sarkozy lost a lot supporters in favors of Fillon. After the second round won by Fillon, the main process occurred and Juppe lost a lot of supporter. Another interesting evolution is the fusion of the community of Sarkozy and Fillon which could be explained by Sarkozy’s choice to support Fillon between the first and second round of the primary.