Personalizing explanations in recommender systems with different levels of details based on users\' reviews

• Ano de defesa: 2022
• Autor(a) principal: Souza, Luan Soares de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Explanations; Explicações; Recommender systems; Sistemas de recomendação; Sumarização de textos; Text summarization
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-10102022-134457/

The amount of data available on the Internet increases the importance of Recommender Systems. However, recommender systems can be seen as a black box to the users, motivating studies to explain why a recommendation has been made. Most of the approaches only use the structured data and the users interactions, and recently, approaches have begun to use the users annotations. Nevertheless, explanation algorithms are frequently limited to a restricted group of recommendation algorithms that can be explained. However, agnostic approaches have been studied to solve this problem. Moreover, proposals of recommender systems with serendipity have become common, and explanations with different levels of details can improve the knowledge discovery process and the users satisfaction. In this context, this project proposes to enable explanations with different levels of detail, utilizing user reviews and hierarchical clustering techniques. The hierarchical clustering enables a vision of different levels of granularity of aspects, and user reviews may provide aspects and sentences to summarize their explanations.