Estratégia e projeto de interação baseados em agrupamento de itens para a elicitação de preferências de novos usuários em sistemas de recomendação

• Ano de defesa: 2021
• Autor(a) principal: Soares, Fabrício Leite
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Informática; Programa de Pós-Graduação em Informática; UFPB
• Palavras-chave em Português: Informática; Aprendizagem de máquina; Problema da inicialização fria; Filtragem colaborativa; Clusterização; Interação homem-máquina; Sistemas de recomendação; Computing; Machine learning; Cold start problem; Collaborative filtering; Clustering; Human–computer interaction; Recommendation systems; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/26329

In order to provide tailored recommendations to a user, a Recommendation System makes use of a preference elicitation process whenever a new user is registered in the system. We propose that this task is best achieved not by the classic method, in which users first express their preferences for individual items, but instead by expressing preferences for groups of items, since the classic method inefficiently converts a user’s effort into a personalized profile. We tested this idea by developing and evaluating an interactive process, where users express preferences through groups of items that are automatically generated by clustering algorithms (clustering). This strategy of recommendation can be applied to any collaborative filtering based system. We assess our process, both with offline simulation methods, using the MovieLens data set; and with an online experiment with 312 users. Our evaluation reveals pros and cons associated with moving from preference elicitation per item to preference elicitation per group. The user experiments showed that the top-N recommendations list generated by our proposal contains more itens which users may be interested than the classic method, and that a greater number of groups positively impacts prediction accuracy, at the expense of greater user effort. Furthermore, we have found that, in comparison with a baseline interface of 15 items, users are able to complete the preference elicitation process in less than half the time.