| With the popularization of Internet and the rapid development of E-Commerce, information overload made it hard for consumers to find the products and services they wanted within a mass of product information. To address this issue, recommendation systems were used to suggest products and to provide consumers with information to help them decide which products to purchase.Although recommendation systems have been very successful in both research and practice, they suffer from sparsity and cold-start problems which affect the performance of recommendation badly. Aimed at the main challenges of recommendation systems, this thesis explored and researched the recommendation systems and their key recommendation technologies, especially collaborative filtering algorithm including user-based collaborative filtering algorithm and item-based collaborative filtering algorithm.The main research works in this thesis are as follows:1) Improved SVD-based collaborative filtering algorithm. This method computed the best rank-k approximation matrix which captured the k most prominent features of the data by minimizing the residual error among all rank-k matrices and then used the feature matrices to produce the prediction of unrated items. Our experiment results suggested that this method could uncover latent features of the given data and efficiently overcome the extreme sparsity of user rating data.2) Improved item-based collaborative filtering algorithm. The traditional item-based collaborative filtering algorithm related items by various heuristic variants of correlation coefficients, which allowed direct interpolation from neighbors' scores. This method offered a rigorous alternative to these interpolation weights based on global optimization of a cost function correlated with all weights simultaneously. Our experiment results suggested that this method could overcome the cold-start problems and improve the recommendation quality with a minor increase in running time. |
PDF Full Text Request