Personalized Movie Recommendation System based on Proximal Policy Optimization

In modern era, personalized recommendation systems have become crucial for enhancing user experience by providing personalized content recommendations across various platforms such as social media, e-commerce and streaming services. However, the existing Deep Deterministic Policy Gradient (DDGP) based movie recommendation system faced catastrophic forgetting and struggles in noisy or dynamic user preferences. Hence, to address these limitations, a Proximal Policy Optimization (PPO) based personalized movie recommendation system is proposed to enhance the accuracy and stability of personalized movie recommendations. Initially, user movie interactions data, user profiles and movie meta data is collected from the MovieLens dataset. Next, the acquired data is preprocessed using data cleaning to handle missing values and remove outliers. Furthermore, One Hot Encoding (OHE) is applied to convert categorical features into binary vectors and min max scaling to perform normalization. The recommendation task is modelled as a Markov Decision Process (MDP), where PPO agent interacts with the environment, learns an optimal policy and receive feedbacks. Finally, the PPO is employed to optimize the movie recommendation policy based on user preferences, enabling accurate and stable recommendations. Therefore, the proposed PPO based movie recommendation system outperformed existing DDGP by achieving 81% precision, 85% recall and 83% F1-score respectively.