Bojana Soro

User Needs models are the backbone of the contemporary audience-focused editorial strategy. These qualitative models identify specific user needs (e.g. “update me”, “keep me on trend”, “inspire me”, “educate me”) that are then used as both inputs and criteria in editorial decisions. User needs are thus seen as quite important drivers of digital media content production. Various newsrooms have postulated different User Needs models in accordance with their specific audiences and focuses, but most of them revolve around the initial research of Dmitry Shishkin in BBC and the form of the User Needs model that we will refer to as “the canonical model” in this talk. Given the importance of the User Needs models for the editorial strategy, the question of (a) the operationalization of this conceptual model and (b) research on the possibility of automatic detection of user needs via ML are of immediate interest in Data Science. Smartocto is currently at the forefront of such efforts in the editorial analytics arena. We will present and discuss our initial approach to feature engineering and predictive modeling of canonical user needs from news articles. Constrained by the overly insufficient sample sizes in these initial studies, we were already able to show promising results both in (1) the automatic detection of user needs from news articles and (2) prediction of the overall article performance based on user needs.