Building the Twitter Network.

People from the same faction tend to retweet each other as a form of social endorsement, as documented in the relevant literature (23), while cross-faction retweets are less likely. Considering that, only retweets would pose the question of how to get rid of spurious or infrequent interactions, possibly by identifying a given retweet threshold. Identifying a threshold would be problematic, as the final network structure might greatly vary with small perturbations on the considered threshold, as can happen on co-occurrence networks (34). We address this issue by considering strong social interactions: Twitter interactions where users perform at least one retweet but also at least another type of Twitter interaction, be it a mention or a reply during the considered time window. Notice that mentions and replies do not express the same social endorsement of retweets, but they can help in identifying the core interactions in the considered social system. The resulting Twitter Core Network (TCN) included 12,000 users, and 16,000 directed strong social interactions. Notice that the TCN aggregates interactions happening over the whole considered time window. However, the frequency of Twitter interactions strongly correlates with the indegree on the TCN (Kendall tau τ=0.81), thus indicating that the aggregated network topology is a valid proxy for investigating patterns of Twitter interactions.