Topic Modeling for Exploring Cancer-Related Coverage in Journalistic Texts

Abstract

Topic modeling has been used for many applications, but has not been applied to science and health communication research yet. In this paper, using topic modeling for this novel domain is explored, by investigating the coverage of cancer in news items from the New York Times since 1970 with the Latent Dirichlet Allocation (LDA) model. Content analysis of cancer in print media has been performed before, but at a much smaller scope and with manual rather than computational analysis. We collected 45.684 articles concerning cancer via the New York Times API to build the LDA model upon. Our results show a predominance of breast cancer in news articles as compared with other types of cancer, similar to previous studies. Additionally, our topic model shows 6 distinct topics: research on cancer, lifestyle and mortality, the healthcare system, business and insurance issues regarding cancer treatment, environmental politics and American politics on cancer-related policies. Since topic modeling is a computational technique, the model has more difficulty with understanding the meaning of the analyzed text than (most) humans. Therefore, future research will be set up to let the public contribute to analysis of a topic model.