AWSGLD Algorithm for Video Content Recommendation System

The OTT service market has rapidly expanded with the commercialization of smartphones and the COVID-19 pandemic, restructuring the media industry and establishing itself as a major form of leisure. In this competitive landscape, personalized recommendation systems play a crucial role in enhancing user experience and service loyalty by recommending content based on individual preferences. BPR, a widely used recommendation algorithm, estimates personalized rankings using SGD. Although simple and easy to implement, SGD struggles with the exploration of anisotropic functions, limiting its efficiency in high-dimensional data analysis. SGLD was developed to improve sampling efficiency and convergence to the posterior distribution, but it remains prone to local traps. To address this limitation, AWSGLD was proposed to facilitate global optimization by adaptively adjusting energy distributions during simulations. This study applies AWSGLD to BPR estimation using Netflix and MovieLens rating data. The results demonstrate that AWSGLD outperforms both SGD and SGLD in recommendation accuracy and ranking quality.