access_time 23 de janeiro de 2020 às 14:00 até 23 de janeiro de 2020 às 15:00
place Room 2.N7.1, IST, Taguspark
While typically recommender systems work on a user-item basis, people-to-people recommender systems focus on recommending users to other users. We believe that these systems might posit societal benefits by helping users establish positive relations with others. Here, using a structured and systematized review procedure, we aim to shed light on the primary challenges, trends and advancements in the field over the past 10 years, with a special focus on people-to-people recommendation systems.We start with a higher-level systematic mapping study, to identify the most relevant topics in recommender systems. 1371 papers were subject to title and abstract screening, resulting in a final collection of 302 papers reviewed by means of a full-text analysis. We detected a preference towards collaborative filtering and identified dimensionality reduction, data sparsity, cold-start and similarity as the main problems addressed in the field. Then a narrative synthesis was conducted, pointing at matrix factorization and deep learning solutions as the most stand-out approaches in the literature. Finally, we provide a closer look at the particularities of people-to-people recommendations, where we find that implicit preferences outperform explicit preferences, and that people are more accepting of recommendations from friends and people they trust.
local_offer Tópicos de Investigação
person Candidato: Pedro Marques do Carmo Rodrigues
supervisor_account Orientador: Prof. João Miguel De Sousa de Assis Dias / Prof. Rui Filipe Fernandes Prada