Collaborative filtering (CF) is one of the most popular recommender system technologies, and utilizes the known preferences of a group of users to predict the unknown preference of a new user. However, the existing CF techniques has the drawback that it requires the entire existing data be maintained and analyzed repeatedly whenever new user ratings are added. To avoid such a problem and improve computational efficiency, this thesis develops three dimensionality reduction algorithms for CF.
To overcome the disadvantage of the existing CF techniques, a new approach called Eigentaste was proposed based on the principal component analysis (PCA). However, Eigentaste requires that each user rate every item in the so called gauge set for executing PCA, which may not be always feasible in practice. Developed in the first study in this thesis is an iterative PCA approach in which no gauge set is required. The developed approach simultaneously estimates the missing values and determines the PC’s using SVD. The PC values of users in the reduced dimension are then used for clustering users. The developed approach and Eigentaste, combined with two clustering methods, are compared in terms of the mean absolute error (MAE) of prediction using three real data sets. Computational results indicate that the prediction accuracy of the proposed approach does not deteriorate even without a gauge set, and therefore, the proposed approach may be considered as a useful alternative when it is neither possible nor practical to define a gauge set.
The iterative PCA approach using SVD takes a considerable amount of time and space to estimate a new user’s missing ratings. To alleviate this problem, two SVD update methods, the Zha and Simon and the folding-in methods, are considered as possible alternatives in the second study. These alternatives are compared in terms of both the MAE and computational time using a real data set. The experimental results show that the SVD update method b...