Profiling Facebook users political preferences

Social media has become an integral part of modern political campaigning because it offers a cost-effective way for politicians to communicate their message with a huge amount of potential voters. We have already seen many examples of social media leveraging for political campaigns, none more effectively than Cambridge Analytica’s work with the Trump campaign in 2016. By combining the vast amount of voter information available in the US with their psychological profiling methodology, Cambridge Analytica was able to compile granular mapping of individual preferences and behaviours. This helped the Trump campaign to tailor its message to an almost individual level, resulting in an electoral win that many believed was not possible within the country’s established democratic norms. But can this approach be replicated in situations/countries where detailed voter information is not available as in the US?

This question was central to a recent QI project with a senior politician in an Asian country. To compensate for the lack of detailed voter information available in the client’s country, QI formulated a methodology to identify undecided voters on social media that the client can subsequently use to more effectively communicate his messages online. Working with the client’s team of political analysts and their knowledge of the local political landscape, we initially identified profiles for supporters of the three main political blocs in the client’s country along with distinct profiles of undecided voters. We then devised a collection strategy targeting Facebook pages that were overtly political, such as those associated with political parties, anti- or pro-government media outlets, civil society organizations and informal and formal pressure groups.

Next, QI developed an algorithm that automatically profiled users interacting with the Facebook pages based on their overall behaviour. We studied which pages they interact with, what reaction they had to each page and what posts thematic they reacted the most. Using a mix of rule-based algorithms and machine learning, we quickly identify an initial set of committed and swing voters. Focusing on the swing voters only, a manual audit of the initial results produced very encouraging results with over 60% accuracy. As we add more Facebook pages to the process and refine the algorithm rules, the accuracy rate will increase, giving the client a list of qualified undecided voters to which he will be able to deliver appropriate campaign messaging.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.