Context Matters: Time as a Parameter in Explorable Hierarchical Topic Modeling in Large-Scale Cross-Platform Online Discourses
- Felix Victor Münch (Leibniz-Institut für Medienforschung | Hans-Bredow-Institut)
Abstract
In times of several global crises at once, such as climate change, pandemics, and armed conflicts with global repercussions it seems a daunting task to gain a general overview of the most relevant parts of public discourse. Superficially unrelated discourses influence each other on multiple levels of scopes and timescales. Therefore, this endeavour necessitates approaches that can simplify the inherent complexity of these topics at a top-level while allowing data analysts to choose where to dig deeper. Not only the interactive explorability of topic relations and sub-topic structures is key, but ever-changing topics and crises necessitate the introduction of time as a topic-defining parameter. Therefore, we developed a combination of methods, leaning on the embedding-clustering sequence of BERTopic, but based on semantic similarity networks, augmented with a time-parameter, and hierarchical network community detection methods. In this contribution we describe our method and its validation by means of a large-scale test case, namely the German-language cross-platform discourse around climate change on Facebook, Instagram, Twitter, and Telegram from 2019 to 2023. Furthermore, we present how the method supported analysts in discovering known and novel dis- and misinformation narratives. Lastly, we will discuss how not only time but also other contextual data, such as the behavioural history of content producers, could augment and improve the results of this and similar natural language processing methods in the future.