(What) Can Journalism Studies Learn from Supervised Machine Learning?

ABSTRACT

In recent years, scholars have explored the applicability of supervised machine learning (SML) within journalism studies. While such computational methods could be of added value to the field, the rationale for employing these supervised models harbors some assumptions that deserve further inspection. This paper seeks to specify under which conditions SML could be useful for journalism scholars and where the field stands in exploiting its potential benefits. We start with an introduction to SML and give an overview of its applications within journalism studies. Next, we identify challenges for the field in its adoption of such techniques. These include overstating the time and financial savings caused by automatic coding, neglecting proper sampling methods, the danger of algorithmic determinism and the limited generalizability of predictive modeling across different domains, contexts and time periods. At the same time, we distinguish several opportunities. These include sharing classifiers, standardizing coding schemes and adopting general purpose techniques. Most importantly, in order for SML to contribute to the epistemological advancements in the field, SML could be used to explain how long-standing theories in journalism are changing. In turn, this might help us to disentangle the inner workings of our contemporary complex news ecosystem.

KEYWORDS:

• supervised machine learning
• computational methods
• big data
• content analysis
• digital methods
• predictive modeling

Disclosure Statement

No potential conflict of interest was reported by the author(s).

ORCID

Lieven De Marez http://orcid.org/0000-0001-7716-4079