Optimizing Recommender Systems with TS Algorithm: Personalized Adjustment through Utilizing User Ratings Offset

This paper discusses the evolution of recommendation systems, focusing on the multi-armed bandit algorithm's role in optimizing user content connections. It traces the algorithm's history from its inception in 1952 for investment problems to its current application in AI-based recommendation systems. The paper emphasizes the importance of balancing exploration and exploitation in recommendation systems, highlighting the Markov Decision Process framework. It delves into the specifics of the Thompson Sampling (TS) algorithm, which utilizes a Bayesian approach for better uncertainty modeling. An advanced version, Offset Thompson Sampling (OTS), is proposed to address the limitations of traditional TS by incorporating user feedback and preferences. The paper also presents a practical application of OTS using the MovieLens dataset, demonstrating how the algorithm adapts to dynamic changes in user preferences. The study illustrates the significance of personalized recommendation systems in today's digital landscape and their continuous evolution to enhance user experience.