Data clustering based on prediction regression models: developments and applications in recommender systems.

Recommender Systems (RS) are powerful and popular tools for e-commerce. To build its recommendations, RS make use of multiple data sources, capture the characteristics of items, users and their transactions, and take advantage of prediction models. Given the large amount of data involved in the predictions made by RS, is unlikely that all predictions can be well represented by a single global model. Another important aspect to note is the problem known as cold-start that, despite that recent advances in the RS area, it is still a relevant issue that deserves further attention. The problem arises due to the lack of prior information about new users and new items. This thesis presents a hybrid recommendation approach that addresses the (pure) cold start problem, where no collaborative information (ratings) is available for new users. The approach is based on an existing algorithm, named SCOAL (Simultaneous Co-Clustering and Learning). In its original version, based on multiple linear prediction models, the SCOAL algorithm has shown to be efficient and versatile. In addition, it can be used in a wide range of problems of classification and / or regression. The SCOAL algorithm showed impressive results with the use of linear prediction models, but there is still room for improvements with nonlinear models. From this perspective, this thesis presents a variant of the SCOAL based on Extreme Learning Machines. Besides improving the accuracy, another important issue related to the development of RS is system scalability. In this sense, a parallel version of the SCOAL, based on OpenMP, was developed, aimed at minimizing the computational cost involved as prediction models are learned. Experiments using real-world datasets has shown that all proposed developments make SCOAL algorithm even more attractive for a variety of practical applications. Key-words: Recommender Systems, Cold-start, Biclustering, Prediction models, Extreme Learning Machines. LISTA DE ILUSTRAÇÕES Figura 1 – Comparação das avaliações feitas pelos Usuários 1,2 e 5. . . . . . . . . . . 29 Figura 2 – Dados utilizados pelo SCOAL para a construção dos modelos de predição — adaptado de Deodhar e Ghosh (2007). . . . . . . . . . . . . . . . . . . . . 41 Figura 3 – (a) Solução encontrada pelo SCOAL depois de rearranjar as linhas e colunas da Tabela 1. Linhas e colunas são realocadas para outro bicluster sempre que tal operação possibilite uma melhoria na média de desempenho dos modelos de predição. As diferentes cores representam os biclusters induzidos. (b) Mapeamento de entrada-saída produzido pelo modelo de predição {r,c}. Os índices de linha e coluna são específicos para cada bicluster. . . . . . . . . . 42 Figura 4 – Visão geral da abordagem proposta para lidar com o problema cold-start. . . 48 Figura 5 – Valores de NMAE para diferentes números de avaliações disponíveis — situações de cold-start parcial. . . . . . . . . . . . . . . . . . . . . . . . . 54 Figura 6 – Valores de NMAE para situações extremas de cold-start. . . . . . . . . . . 55 Figura 7 – Acurácia da classificação obtida considerando que cada rótulo de cluster encontrado pelo SCOAL é um rótulo de classe para o problema cold-start. . 56 Figura 8 – NMAE obtido na base Movielens para MK-KNN, NB, J48, Logistic, WP e DC. 57 Figura 9 – Arquitetura de uma rede SHLFNN. . . . . . . . . . . . . . . . . . . . . . . 60 Figura 10 – Saída do algoritmo SCOAL-ELM com U =V = 2. . . . . . . . . . . . . . 63 Figura 11 – Valores do Erro Médio Absoluto Normalizado (NMAE). . . . . . . . . . . . 66 Figura 12 – Valores de precisão. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 Figura 13 – Base Netflix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 Figura 14 – Base Jester . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 Figura 15 – Base Movielens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 Figura 16 – Base Música . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 Figura 17 – Tempos computacionais obtidos pelo SCOAL-LR com o offload automático da biblioteca MKL habilitado e desabilitado. . . . . . . . . . . . . . . . . . 91