Research question/goal:

Social science research increasingly draws on “digital behaviour” data, such as participants’ browsing histories, to complement traditional surveys. Digital behaviour data may be used, for example, to track individuals’ political news consumption online in addition to their self-reported responses to political questions in surveys. However, past research has not fully exploited the rich potential of browser history data but rather extracted ad-hoc measures to infer what political messages participants receive. Meanwhile, Large Language Models (LLMs) have proven effective at various data processing tasks and offer exciting potentials when it comes to multi-modal content (e.g., text, images and videos on websites). This project adapts LLMs to analyze and condense browsing histories with the aim of offering deeper insight into online behaviour. The project thereby enables a new look at the effect of political messaging content on political attitudes and party support.