Tensor Factorization Approach for Real-Time Personalized Content Recommendation in OTT Platforms
Annotatsiya
Over-the-top (OTT) platforms are increasingly leveraging intelligent recommendation systems to personalize user experiences based on viewing behavior, content preferences, and contextual factors. Tensor factorization offers a powerful approach to modeling complex interactions among users, content, and time, thereby enabling multidimensional personalization. However, existing recommendation methods often rely on static or batch-learning models that struggle to adapt to real-time user behavior and dynamic content updates. These limitations result in latency, reduced relevance of recommendations, and suboptimal user engagement. To address these challenges, this study proposes a novel framework called Real-Time Tensor Factorization with Online Stochastic Gradient Descent (RTTF-OSGD). The framework decomposes user-content-time interaction tensors using a lightweight, online learning approach, allowing for continuous updates as new data arrives without requiring retraining of the entire model. The proposed method is applied to generate personalized movie and series recommendations on OTT platforms, adapting in real-time to shifts in user behavior, trending content, and contextual changes such as time of day or device used. Experimental results show that RTTF-OSGD significantly improves recommendation accuracy by 99.34%, user satisfaction by 97.65%, and system responsiveness by 98.71%, as well as scalability by 96.47%, compared to traditional matrix or offline tensor factorization methods, making it ideal for scalable and adaptive content delivery.