Social media discussions have been shown to influence public perceptions on a myriad of topics, including support for health policies and screening programs, and mistrust of government and medical institutions. People often go to social media for health purposes, because they feel uncertain and overloaded by the vast amount of information that is available about cancer and screening. At present, very little is known about the causes and the consequences of these social media discussions on support for the cancer screening program, the public’s trust in national health institutes, their perception of balanced information provision by those health institutes, and their subsequent news media selection.
This project is the first to investigate longitudinally how news media content, social media discussions, and public perceptions towards screening participation are related. In a unique consortium blending key academic expertise on news media effects, public opinion, social media network studies, artificial intelligence, text mining, and health communication, with important societal partners from cancer screening in the Netherlands, we will not only investigate how informed decision making regarding screening and trust in the national screening program is affected by – and affects – news media exposure and social media conversations, but also will develop, validate, and implement technological tools for automatic coding of news and social media content at these societal partners. We will also present a white paper on effective social media strategies to promote informed decision making for screening and trust in public health organizations. This white paper is based on social network simulation models from an extensive and existing data collection of news and social media in the period 2010-2020, and from a new data collection.
The project is divided in various sub-projects.
In sub-project 1, we will develop and validate a codebook for media content analysis, that will serve as the input for training a machine-learning classifier that is able to automatically code content and sentiment of each news media article. We will distill the news media from the database NexisUni in a source file. Identifying which topics and sentiments to code will be done through an iterative process of four strategies: (1) a literature search, (2) existing codebooks on similar topics available in the consortium, (3) focus group interviews with members of the public, and (4) unsupervised machine learning. A consensus on the topics and sentiments will be finalized in a codebook. We will subsequently develop and train machine learning classifiers that are able to automatically code news media content and sentiment.
In sub-project 2, we will adopt a similar approach, but with the ultimate goal to develop and train a machine-learning classifier for social media content and sentiment. Social media content differs from news media content, by being confined to a maximum numbers of characters, and where people use more layman terms and slang. To accommodate for this, we will adapt existing codebooks available in the consortium on social media content and sentiment analysis. As per sub-project 1, we will develop and train machine learning classifiers that are able to automatically code social media content and sentiment.
In sub-project 3, we will manually code network structures of social media platforms to identify important network characteristics in those groups, including nodes in the network and tie strength between group members. In this sub-project, we will work actively together with our societal partners that have social media platforms for cancer (www.kanker.nl), or are active participations on social media themselves (National Institute for Public Health and the Environment; Dutch Cancer Foundation) to identify relevant social media platforms. We will also focus on social media platforms of local and national news websites. The main outcome of sub-project 3 is a representation of diverse social networks in the field of cancer screening.
In sub-project 4, we will perform advanced time-series analysis to understand the dynamics between news media content, social media content, and public perceptions towards screening and screening institutions. The main outcome of this subproject is an understanding of the causes and consequences of social media discussions on cancer screening.
In sub-project 5, we will use agent-based modeling to simulate social media network interventions based on the data collected in the project. The main goal of this sub-project is to identify which message strategies are most likely to promote trust in cancer screening, cancer screening institutions, and balanced information provision to enhance informed decision making with regard to screening.