Public opinion formation on social media in a big data perspective : measuring political homophily and cross-cutting agreement (2020)
Type of ContentTheses / Dissertations
Thesis DisciplineMedia and Communication
Degree NameDoctor of Philosophy
PublisherUniversity of Canterbury
The establishment and constant development of social media technologies over the last two decades have left media scholars with a burning desire to understand their contribution to the formation of public opinion and its political consequences. At this very moment it is undeniable that public opinion fuelled by social media is a force to be reckoned with. At the same time, the complex information flows and communication practices on social media make it difficult to predict how public opinion will form. Research into how information and opinions are shared has created concerns that some parts of the public are becoming increasingly politically homogenous and polarized due to social media.
This thesis project suggests a new framework to evaluate the function of public opinion, specifically in relation to the opportunities offered by social media. The framework is inclusive and conflates concepts that are usually considered separate within the field of media studies e.g. opinion polling and deliberation. It is a descriptive framework that highlights advantages and weaknesses for certain instances of public opinion. The framework is used to inform the development of computational methods that can be used to gauge public opinion with respect to political homophily and polarization. These methods rely on automatically calculating the cross-cutting agreement, i.e. how much agreement there is between people with opposing political affiliations, as they interact in public social media discussions. They are developed as universal methods to enable public opinion to be measured based on how politically homogenous the users who react to and share a piece of information are and how well political oppositions in the comment threads are able to reach agreement, in combination with topics and sentiments. The methods are tested on a large cross-section of public Danish Facebook pages related to everything from local fitness clubs to government pages and news media organization.
Results show that high levels of political homophily and disagreement that are likely to cause polarization are highly dependent on context. Overall polarization does not appear to be increasing over a period of five years, except if moderated by certain other factors of which discussion topics related to refugees and immigration is one of the strongest. The results are furthermore an indicator that the methods developed can be used to effectively evaluate public opinion with respect to homophily and polarization. Since the methods are computational, the process is easily automated and can be used to create