The Epistemology of Fine-Grained News Classification

The process of news digitalization over the past decades has released massive amounts of news content, revolutionizing consumer access to news and disrupting traditional business models. These radical changes have also introduced new opportunities for media content analysis, opening up new scenarios for ambitious large-scale media analytics initiatives, which can go well beyond the relatively small-scale studies currently carried out by media scholars and practitioners. However, take-up of computational methods to support media content analysis activities has been rather modest, reflecting a degree of disconnect between the needs of scholars and practitioners for task-specific and usable software solutions and the state of the art in computational techniques for news media analysis. In this article, we perform an initial step toward bridging this gap, by looking in detail at the task of fine-grained news classification. In particular, we propose a typology of news topics, which is formally specified and realized into a family of reusable ontologies. The proposed model has been validated empirically, through an analysis of a multilingual news corpus, as well as formally, in terms of the functional and logical properties of the ontologies. Our analysis brings together the media and computer science literature, connecting the formal definitions provided in this article to the concepts used by media scholars.