Addressing Initialization and Data Ordering Issues in Latent Factor-Based Recommendation Systems

Recommendation systems play a crucial role in helping users navigate information overload, particularly in today's digital era. Their primary objective is to predict users’ preferences for items. Latent factor-based recommendation systems achieve this by aligning users and items under latent factors. Previous studies mainly focused on devising effective objective functions for learning these latent factors. However, the accuracy of latent factors also depends on their initialization and the order of the collected data fed into the training. Therefore, in this paper, we propose methods to address these two issues. The experiments were conducted on two standard datasets, Movielens 1M and Yahoo Webscope R4, using the RMSE metric. The experimental results indicated that our proposed methods improve the accuracy of latent factor-based recommendation systems.