Unveiling Political Influence Through Social Media: Network and Causal Dynamics in the 2022 French Presidential Election

TL;DR

This paper develops a data-driven framework to map political influence on Twitter during the 2022 French presidential election by combining time-series analysis of 11 topics across 7 political communities with network centrality and nonlinear causal inference. It demonstrates how collective dynamics emerge prior to key electoral moments, identifies central actors and topics through correlation-based graphs, and uncovers directional, asymmetric influence patterns using Convergent Cross Mapping (CCM). The study highlights health and foreign policy as pivotal catalysts for cross-party influence and reveals how the post-election landscape reorganizes around new alliances and discourse dynamics. The approach provides a practical framework for real-time monitoring of political discourse and informs campaign strategists and media analysts about evolving influence pathways.

Abstract

During the 2022 French presidential election, we collected daily Twitter messages on key topics posted by political candidates and their close networks. Using a data-driven approach, we analyze interactions among political parties, identifying central topics that shape the landscape of political debate. Moving beyond traditional correlation analyses, we apply a causal inference technique: Convergent Cross Mapping, to uncover directional influences among political communities, revealing how some parties are more likely to initiate changes in activity while others tend to respond. This approach allows us to distinguish true influence from mere correlation, highlighting asymmetric relationships and hidden dynamics within the social media political network. Our findings demonstrate how specific issues, such as health and foreign policy, act as catalysts for cross-party influence, particularly during critical election phases. These insights provide a novel framework for understanding political discourse dynamics and have practical implications for campaign strategists and media analysts seeking to monitor and respond to shifts in political influence in real time.

Figures (10)

• Figure 1: Political twittersphere over the period from 2021-09-01 to 2022-12-31. Following gaumont_reconstruction_2018. Each node is a Twitter account, each color correspond to a political digital community. Political communities have been reconstructed from the retweets graph were links represent the retweet activity over the full period. Louvain clustering blondel_fast_2008 has been used to detect the communities and the graph has been spatialized with the Force Atlas algorithm jacomy2014forceatlas2. Only links with weight highers than 20 retweets have been kept for the community detection and only links corresponding to 100 retweets or more are shown. Moreover, only the communities associated to political leaders studied in this paper are displayed on this image, which represent 39.4k accounts (80 % of the full map).
• Figure 2: Time series of the daily number of tweets
• Figure 3: Time series of log-return of number of tweets (change of activities). Each time series is shifted to avoid overlapping and regrouped by topics for clarity (they all have a mean of zero).
• Figure 4: Top panel: Evolution of the largest eigenvalues relative to those computed over the entire period. Bottom panel: Topics corresponding to the highest degree, annotated with key political events.
• Figure 5: Correlation network structure pre-election