Framework for Movie Recommendation System using GNN and Textual Data

The surge in on-demand movie platforms and the constant influx of new film releases presents a significant challenge in effectively recommending content to users. With a vast array of movies uploaded daily, online platforms struggle to host diverse content and ensure tailored discovery for users based on their preferences. The platform’s effectiveness relies heavily on recommending content aligned with user interests, making recommendation engines crucial for user retention and overall success. These systems typically involve creating algorithms to learn user and movie features and provide recommendations. Despite the wealth of user-related data, obtaining accurate and timely movie information, especially for newly introduced films, remains complex. The absence of historical data for recent movies complicates accurate recommendations. Movie subtitles, containing comprehensive dialogues, can serve as a valuable information source for recommendation engines. This paper leverages the growing popularity of Graph Neural Networks (GNNs) in recommendation systems to capitalize on their ability to discern intricate relationships between entities such as movies, users, and actors. Graph structures naturally represent relationships between entities, allowing the explicit modeling of complex relationships, dependencies, and interactions. By combining subtitle analysis with graph-based learning, the proposed system addresses the challenge of recommending movies without relying on historical user interaction data.