Effects of Headline Formulation Features

The aim of the project was twofold: on the one hand, we wanted to verify whether formulations of headlines play a role in news selection decisions of news users, and whether social media algorithms would help reduce the diversity of social media algorithms (that is, do algorithms favor news that is formulated in specific ways?). In order to achieve these goals, we replicated and extended a content analysis of headlines with Click Through Rates (CTR) collected in A/B tests by the Dutch newspaper NRC, and we further developed a tool simulating a news recommender, in order to tap audience preferences for news presented in different formulation styles.

The first stage of the research project investigated formulation features of NRC newspaper headlines that had been modified in so called A/B tests: alternative headlines were probed on identical website pages, where the difference in the number of clicks decided which was the more popular headline. The tests were performed without any presumptions about our research; we obtained the data set of about 8,000 headlines with their views and clicks afterwards. We wanted to verify whether properties like negativity, language intensity, and forward referencing would promote headline selection. In a content analysis these and other features were identified and tested against the CTR. This way we replicated and extended earlier research (Lagerwerf and Govaert, 2021). A manuscript to be submitted to Digital Journalism  is almost completed.

The second stage of the research project consisted of the development of a digital tool (a recommender systems research tool called ‘3bij3’, Loecherbach & Trilling, 2020) to further investigate potential systematic influence of the formulation features negativity and language intensity. First, news items gathered from news website were clustered on the basis of their news topic. Second, from a set of similar news items two headlines were chosen that are opposite in negativity. Likewise, two headlines were chosen that are opposite in language intensity. These headlines are, together with two filler items, presented on a mock news website. Every day, news items are renewed, and the same respondents are asked to pick the news item that interests them the most. Their choices are registered. In our data analysis we are analyzing systematic patterns of news choices motivated by negativity and/or language intensity. The tool has been developed and data collection has started this week.

Lagerwerf, L., & Govaert, C. G. (2021). Raising clickworthiness: Effects of foregrounding news values in online newspaper headlines News Values from an Audience Perspective (pp. 95-119): Springer.
Loecherbach, F. & Trilling, D. (2020). 3bij3 – Developing a framework for researching recommender systems and their effects. Computational Communication Research 2(1), 53-79. doi: 10.5117/CCR2020.1.003.LOEC

Academy Assistants

Supervisors: Luuk Lagerwerf & Wouter Atteveldt
Advisors: Nicolas Mattis and Tim Groot Kormelink