Designing an Algorithmic News Application for User Control

• Autor(a) principal: Tiago Nuno Mesquita Folgado Leitão Devezas
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: https://hdl.handle.net/10216/105729
• topic: Outras ciências da engenharia e tecnologias; Other engineering and technologies

The torrential stream of information published online every day has fostered the rise of technological solutions aiming to combat information overload. News aggregators, applications that collect news from multiple publishers and present them in a condensed form in a single place, are one such solution. In order to increase efficiency, many aggregators now employ personalization and curation algorithms, computer programs that decide what to present based on a set of rules and criteria. But this approach can entail new problems. Algorithms can be biased, lock users inside "filter bubbles", and help disseminate false news. This power to shape public opinion is often left unchecked. Information about the algorithm's inner workings isn't provided to users, who sometimes aren't even aware that their news feed is managed by a computer program. This study aims to identify the importance of a set of criteria that a sample of potential users would like to manipulate in algorithmically generated news aggregator, assess the importance of providing information about the algorithm, and determine the desired degree of control over the system. The goal is to then apply the most relevant findings to a functional prototype that fits the requirements expressed by the users. In order to achieve these objectives, a quantitative non-experimental research design was employed. A questionnaire was developed and sent via email to the University of Porto community. Four hundred and thirty two (432) participants, mainly comprising students and professors, filled the questionnaire. Data was collected about demographics, news consumption habits, and the importance level of multiple criteria regarding news content, diversity of perspectives, news sources, information about the algorithm, and the degree of personalization. Criteria importance level was rated using a five-item Likert scale with the following value-label pairings: 1 - "Not at all important", 2 - "Little important", 3 - "Moderately important", 4 - "Important", and 5 - "Very important". Results indicate that, with a median level of importance of 4, the most important news content criteria were "Timeliness", "Good news", "Proximity", "Conflict", "Serendipity", "Magnitude", "Relevance", and "Audio-visual". Regarding the diversity of perspectives, participants attributed higher importance to news that present points of view diverse from their own (median = 4) than to content aligned with their views. Professionalism (median= 5) and journalistic prestige (median = 4) were rated as the most important factors concerning the sources. Information about the algorithm was evaluated as being of moderated importance (median = 3). As for personalization, participants indicated that they value the most being able to configure the system according to their own criteria (median = 4), rather than have no personalization or letting the system make those choices. This study provides some insights about the importance of several criteria for manipulating an algorithmic news application, from a user perspective. Some of the findings have already been implemented in a functional prototype, which will be further developed in the future. Evaluating the system with users is also planned as future work. The usage of a non-probability sampling method limits the generalization of the findings, but we believe they can be a valuable reference for implementers of algorithmic news applications concerned with user control and transparency.