Aplicação de rede neural sem peso e locality-sensitive hashing para o problema de cold-start item em filtragem colaborativa

• Ano de defesa: 2018
• Autor(a) principal: Cotta, Kleyton Pontes
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Rio de Janeiro; Brasil; Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia; Programa de Pós-Graduação em Engenharia de Sistemas e Computação; UFRJ
• Palavras-chave em Português: Sistema de recomendação; Arranque a frio; CNPQ::ENGENHARIAS
• Link de acesso: http://hdl.handle.net/11422/12957

Recommender systems are techniques and tools used to suggest customized items based on the user profile. Collaborative filtering is one of the most successful approaches in recommender systems. This approach aims to recommend an item to a user based on items previously evaluated by other system users. However, it is widely known that approaches to collaborative filtering methods suffer from problems such as cold-start, scalability and scatter. The cold-start problem is a longstanding dilemma in recommendation systems. Occurs when there is unavailability of adequate information about the items or users available in the system and therefore it is not possible to make relevant recommendations. In this context, the paper presents a hybrid approach to the problem of cold-start items in collaborative filtering for recommender systems. We used the Locality-Sensitive Hashing (LSH) technique with the MinMaxwise method for word processing in order to find similarities among the items. LSH was incorporated into traditional literature approaches and other machine learning methods, generating good results for these methods. The WiSARD obtained better results regarding training time and for bases with the presence of items with complete cold-start.