Abstract

In contemporary hyper-connected societies, a peculiar paradox characterizes the collective representations gravitating in the social imaginary. On the one side, they rapidly spread and evolve through communication on social media; on the other side, they are objectified in the form of digital data organized in textual databases, persistent and searchable. This article aims to present a quantitative text analysis technique known as «topic modeling» - which allows the fast exploration of «big» text data while taking into account the polysemic and relational character of language, thus fostering an interpretive gaze. Topic models are increasingly employed in the fields of digital humanities, political sciences and cultural sociology. Here, I will illustrate the methodological implications of topic models using a non-technical language and focusing, in particular, on applicability to online communicative interactions. I will present a case study consisting in the analysis of about 420k unique tweets regarding the 2016 edition of Festival di Sanremo. Through topic modeling, I inductively reconstructed the main frames employed by more than 88k users twitting about this media event. Subsequently, I was able to automatically identify the prevalent frame adopted by each Twitter user involved in the digital discussion.