A data science approach to social science problems: examining political bias in false information on social media
The 2016 United States (U.S) presidential election highlighted the powerful influence that social media can have on politics. Fake news stories shared on social media are argued to have swayed the outcome of the last U.S. election and a recent article in the Guardian questioned: “Will fake news wreck the coming [UK] general election?”
With the Conservatives recently spending approximately £100,000 and the Brexit party spending £107,000 on Facebook advertising in the UK, it is becoming increasingly important to understand how false information on social media is politically biased.
As part of their role in the EU funded EUNOMIA (user-oriented, secure, trustful & decentralised social media) project, Trilateral’s research team sought to address this question.
EUNOMIA is a three-year (2018-2021) project that brings together 10 partners who will develop a decentralised, open-source solution to assist social media users (traditional media journalists, social journalists and citizen users) in determining the trustworthiness of information.
An interdisciplinary approach
Trilateral leads the work in EUNOMIA to understand the social and political considerations in the verification of social media information. The research team’s interdisciplinary approach drew on the expertise of social scientists from the Applied Research & Innovation team and data scientists from the technical team.
The social science research comprised of desk-based research and 19 semi-structured interviews with citizen social media users, traditional media journalists, social journalists and other relevant stakeholders.
Participant observation has been undertaken through a data science approach designed to examine political bias in relation to the engagement with false information. To do this the study focused on the political leanings of false information accounts in relation to UK political parties. The data science methodology included:
- Web scraping to identify the Twitter handles of the 579 UK MPs and 49 false information organisations which included organisations labelled as “conspiracy-pseudoscience” or “questionable” sources on the Mediabiasfactcheck.com database and climate change denial organisations listed on the DeSmog Climate Disinformation Research Database
- Data mining on Twitter, processing and utilising big data exclusively using the publicly available Twitter APIs
- Social network analysis to track the following between the examined accounts and extract insights to measure political bias
To examine political bias in relation to false information, the technical team analysed three metrics, including:
- The intersection of the followers of each false information account and the followers of the Conservative and Labour MPs
- A social network graph highlighting the political bias of the followers of each false information account
- Whether false information accounts follow more Conservative or Labour MPs
The findings of all three metrics examined indicate that the majority of false information accounts in this study are Conservative-leaning.
Whilst these findings may be the result of the calls by Conservative-leaning politicians to distrust mainstream media or in how accounts are labelled as false information by fact-checkers, the presence of political bias in social media content can result in it being distrusted.
The interview findings highlighted how information and sources that are politically biased or radicalised are not perceived to be trustworthy. As interviewees highlighted how the language used can provide insights into political bias, our further research will draw on natural language processing techniques to explore the language used when engaging with false information accounts.
The findings of this study have been submitted for publication as a journal article.
For more information contact our team:
Su Anson, Research Manager at Trilateral Research
Xuan Li, Senior Data Scientist & Cloud Solutions Architect at Trilateral Research
Graham Hesketh, Head of Data Science at Trilateral Research
Pinelopi Troullinou, Research Analyst at Trilateral Research