Comparison of The Performances of Clustering and Dimensionality Reduction Approaches in Collaborative Filtering

Recommendation systems (RS) can be defined as systems that aim to offer personalized product and service recommendations to users based on users' past product preferences and similarities with other users in the system, especially in systems that provide e-commerce services. The main purpose of RS is to reveal meaningful information from large-scale data to users and to recommend systems that aim to simplify the analysis of user behaviors and product attributes. It is possible to divide the techniques used in RS into two main categories content-based and collaborative filtering (CF) according to the information they receive as input. Content-based recommendation systems focus on analyzing the attributes of items such as articles, movies or music to generate tailored recommendations. CF methods analyze user-generated scores for products and services to identify patterns and preferences. The success of CF techniques hinges on accurately identifying user similarities within large datasets. However, in CF techniques, large-scale data sets consisting of a large number of users and the scores given by users to these products are used. Consequently, identifying user similarities in such extensive datasets poses significant challenges. Two different methods are used to overcome this problem. The first method applies clustering analysis to divide the dataset into smaller subsets (user or product), followed by the application of CF techniques. In the other method, dimensionality reduction is performed on a product (object) basis using Singular Value Decomposition (SVD) and Principal Component Analysis (PCA) methods. Up to now, many studies have been carried out using clustering analysis and variable dimensionality reduction methods Despite extensive research, a thorough comparison of clustering and dimensionality reduction methods on real-world datasets remains unexplored. This study aims to compare the performances of eleven clustering techniques of eleven clustering techniques, four of which are non-hierarchical seven of which are hierarchical clustering algorithms, and two variable dimensionality reduction techniques, consisting of SVD and PCA METHODS, in CF.