Abstract
Even though we live in the golden age of the Internet and social networks, the “idiot box” remains the primary and undisputed source of political information. Meanwhile, the production and consumption of videos, exemplified by platforms like TikTok and Instagram Reels, are becoming ubiquitous. However, social scientists rarely use television and videos as primary data sources in quantitative research due to the costs and technical challenges associated with the data collection process. Recent advancements and proliferation of machine learning tools for computer vision and audio analysis have finally paved the way to extract meaningful data from large video archives at a relatively low cost. We demonstrate how to apply these tools through a case study involving a web-scraped collection of 2000 episodes of two prominent Italian Political Talk Shows (PTS). We show how it is possible to extract information regarding who is speaking, when they are speaking, what they are speaking about, and how they are speaking (tone of voice and facial expression). We provide a descriptive analysis of the trends observed in terms of political representation, topics discussed, and emotional behaviors. Additionally, we matched the extracted data to minute-by-minute TV viewership data from Auditel™ to describe how different settings are related to changes in viewership.
Draft paper soon available
Slides here:
