Rumour Mining
Mia-Marie Hammarlin, Dimitrios Kokkinakis och Fredrik Miegel 

This presentation draws on our recent work within the 4-year (2020–2023) mixed methods project financed by Riksbankens Jubileumsfond with the aim to study vaccine hesitancy in a Swedish context, on- and offline. The research team involves one ethnologist, one sociologist and two language technologists. We will present methods and preliminary results of the language technology analysis of texts from a number of Swedish social media sites that discuss the topic of “vaccine hesitancy”. Particularly, two techniques from language technology are useful for our purposes, namely sentiment analysis and topic modeling. Sentiment analysis is a process that aims to identify, extract and quantify people's (positive or negative) attitudes, sentiments and/or feelings regarding a text, picture or product. Topic modeling aims to find linguistic patterns of underlying topics in (usually very large) bodies of texts or text fragments. The empirical data, some useful pre-processing steps, techniques and tools, as well as results from the automatic analysis, and thoughts for improving the outcome, by e.g. combing the two technologies, will be presented.

Additionally, we will highlight inherent challenges with mixed methods projects. Some are minor and straightforward to manage, others concern essential epistemological, theoretical, and terminological matters, especially if the methods included in the mix come from as vastly different research fields as the humanities and social sciences on the one hand, and the natural and technological ones on the other. Following Flyvbjerg (2001), we argue that one of the main problems for mixed methods projects such as ours is to find out how to make the episteme and techne of the natural and technological side compatible with the phronetic approach of the social and humanistic one, in a common objective for the project.