Enhancing Personalization in Media Streaming Services Through Graph Neural Network

• Published in: Proceedings (International Confernce on Computational Intelligence and Communication Networks) pp. 1208 - 1213
• Main Authors: Changala, Ravindra, Sneha Sri, S. Sandya, Kadyan, Sunil, C, Sasthi Kumar, Vuyyuru, Veera Ankalu, Bala, B Kiran
• Format: Conference Proceeding
• Language: English
• Published: IEEE 22.12.2024
• Subjects: Accuracy; Adaptation models; Computational modeling; Digital media; Dynamic graph; Graph Neural Network; Graph neural networks; Mean square error methods; Media; Media streaming services; Performance evaluation; Personalization; Predictive models; Reliability; Streaming media
• ISSN: 2472-7555
• DOI: 10.1109/CICN63059.2024.10847340

The increasing diversity of user tastes in media services is a problem for traditional algorithms that aim to provide accurate and customised content choices. This research presents a new method for using Graph Neural Networks (GNNs) to improve personalisation in media streaming services. The links between people, material, and their interactions are sometimes difficult for traditional recommendation algorithms to grasp. GNNs, on the other hand, are excellent at simulating intricate connections in graph-structured data. The suggested approach uses GNNs to build an interaction graph between users and the content, capturing the complex relationships between users, viewing patterns, and content attributes. Through the use of GNNs, people are able to effectively train latent representations that contain client preferences and content attributes in a dynamic manner, leading to suggestions that are more precise and tailored. The suggestion accuracy will continuously increase due to the model's capacity to adapt to changing user behaviour. Extensive trials on real-world video streaming datasets validate the effectiveness of the system, exhibiting better performance than traditional techniques. The Bayesian Graph model lags behind with a greater mean square error (MSE) of 1.25, whilst the GNN shows a competitive MSE of 95, showing effective prediction accuracy for user ratings.