- with Bon E., Gibson R., and Greffet F.
Comparing and Contrasting Manual vs Automated Methods for Detecting Political Influencers Online influencers are increasingly important actors in political campaigns. Their large networks of followers and perceived authenticity make them attractive new sources of leverage for candidates seeking to reach voters, particularly those who are less engaged and trusting of mainstream voices. Despite their growing prominence there is considerable debate about how to define an influencer and this is reflected in the methodological approaches to identifying them. In this paper, we advance this debate by applying and assessing the pros and cons of two distinct approaches – manual inductive vs automated deductive - to detecting political influencers. We focus on the 2022 French Presidential election campaign on Twitter. Specifically, the first approach identifies influencers from YouGov survey respondents accounts who consented to share their Twitter account information. Based on the accounts most mentioned, followed, tweeted or retweeted by participants we develop a coding system that manually categorises influential accounts according to their ideology and status as politicians, news media, or ‘other persons of interest’ (OPOIs). The second approach applies statistical methods to an aggregate dataset of political hashtags from the campaign to identify a ‘take off’ point and isolate the actors involved. We evaluate the performance of each approach post-identification by comparing the influencers they identify in terms of their overlap, follower numbers and network reach. Based on the results of this exercise, we reflect on the trade-offs and benefits of both approaches for researchers seeking to study online influencers, in terms of time, quality of data and sample size.
See: @linegar2023large, @mens2024scalingpoliticaltextslarge, @yu2023openclosedsmalllanguage
- Synthetic data
- Build k-factor prototype
- Check social k-factor
- Add file description & project setup