Multi Modal Graph Fusion Framework for Equitable and Sustainable Content Discovery in Digital Streaming Platforms

This paper presents a fusion-based recommendation framework in which several embedding techniques were integrated to improve content discovery in streaming media platforms, while support for sustainable digital infrastructure and equitable information access was also facilitated. TF-IDF vectorization, sentence transformer embeddings, heterogeneous graph-based meta-path representations, and adjacency-based sparse matrix decomposition were combined to produce unified feature representations. An experimental evaluation carried out on the Netflix dataset showed that the fusion strategy achieved a Precision@5 of 0.68, a MAP@10 of 0.72, and an NDCG@10 of 0.85, with these values surpassing individual baseline methods by 6.25 percent, 9.09 percent, and 8.97 percent. Enhanced diversity was additionally demonstrated by the fusion model, supported by an entropy score of 3.42 and a coverage measure of 0.78, indicating improved content variety and reduced imbalance in content exposure. Classification experiments using Random Forest classifiers produced 92 percent accuracy. Through this integrated methodology, challenges related to sustainable digital content delivery, efficient resource utilization, responsible technological advancement, and inclusive information access were effectively addressed, with a slight elaboration highlighting the framework’s strengthened capability for broad user engagement.