A novel approach to movie recommendation systems: overcoming the cold start problem with dynamic sea-horse slimmable network with a contextual attention network

Movie recommendation systems often face the cold start problem, particularly with new users and items, which complicates accurate recommendations. To address this, we introduce a novel approach utilizing a Dynamic Sea-Horse Slimmable Network with a Contextual Attention Network (DSHS-ConAtNet). This method is applied to four MovieLens datasets: Original 100 K, Original 1 M, Resampled 100 K, and Resampled 1 M. The approach involves pre-processing with Anisotropic Gaussian Filtering and a Directionally Truncated First Derivative to manage noise and sparsity. Feature extraction is performed using Scale-Aware Modulation and Transformer-Based methods to capture high-order features. The DSHS-ConAtNet is optimized using the Sea-Horse Optimizer for efficient resource utilization and personalized recommendations. Collaborative user-based filtering further enhances performance by predicting ratings for unrated items based on user similarity. Implemented in Python, the DSHS-ConAtNet model achieves an accuracy of 99% and an RMSE of 0.6836. This comprehensive approach offers a robust solution for overcoming the cold start problem in movie recommendations.