Sistemas de recomendação em sites de notícias: contextualização como proposta para o engajamento de usuários

Abstract

In a context of exponential increase in content production and changes in the news consumption model, content recommendation systems are seen as elements that help in promoting the engagement of users in news sites. However, the low performance of content recommendation systems, especially the low engagement rates identified in accesses from external sources (social networks, search engines and others that already make up most of the accesses). Thus, the purpose of the research is the recommendation of news on the content pages of news sites, with emphasis on Brazilian sites. This thesis presents the theoretical and practical analysis regarding the use of automated recommendation systems in the news context, in order to validate which aspects of these systems influence the engagement indexes. For this, three hypotheses are proposed to be validated from the results of research and practical experiments, namely: 1) Changes in the design of content pages (including layout, usability and / or criteria of content recommendation elements) impact on the metrics of user engagement with the news site; 2) The "landing page" and "traffic source" variables have a relevant influence on the engagement metrics of news sites; 3) The context interferes with the probability of a particular user consuming a news item (regardless of the level of interest in this content). In addition, problems and factors that bring complexity to the use of automated content recommendation systems in the specific context of journalism and news are identified. Based on analyzes of metrics and research on the current scenario of the recommendation on Brazilian news sites, it is proposed to categorize the criteria used in recommending the content of these sites, as well as conceptual scenarios that point to solutions that promote better results - especially in the issue of user engagement - in the use of recommendation systems on news sites.