Enhancing recommender systems by enrichment with pre- processing approaches supported by users\' feedback

• Ano de defesa: 2019
• Autor(a) principal: Costa, Arthur Fortes da
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Abordagens de pré-processamento; Cold-start; Dimensionalidade; Dimensionality; Esparsidade; Interações de usuários; Partida fria; Pre-processing approaches; Recommender systems; Sistemas de recomendação; Sparsity; Users' feedback
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-04022020-154009/

Recommender systems use information about the users preferences to define scores of interests towards items. Regardless of the method, a noticeable problem is that the system is required to compute scores for a large amount of unknown items in the database, even though these items may not be related to a determined user. Besides that, traditional problems, such as sparsity, high dimensionality and cold-start make the prediction task even more difficult. Currently, several works try to deal with these problems, using solutions within the recommendation algorithm itself, which increases the time and computational cost of them. In this doctoral thesis, we propose many pre-processing techniques for recommender systems that reduce and/or enrich the number of unknown user-item pairs the recommender must process to obtain a dataset with more reliable and robust information. Our approaches focus on users feedback, trying to extract tastes and behaviors from each user from the information available in the datasets. We assess the quality of these approaches by applying them into some well-known RS and comparing the results against the same recommenders without our pre-processing step, as well as against other related baselines and state-of-art works. Results show a significant improvement in the accuracy of the recommenders and the reduction of the impact of the traditional recommendation problems.