Model hyperparameters

Latent dimensions (K): 4 Learning rate (η): 0.010 L2 regularization (λ): 0.020 Epochs per frame: 1

Training

Recommend for

Top-N predicted ratings for the selected user, among videos they have not rated yet.

Training stats

Epoch0

Observed ratings—

Matrix sparsity—

Training RMSE—

Global mean μ—

Users (circles) and videos (squares, colored by genre) are projected from the K-dimensional learned factor space onto its top-2 principal directions. As SGD minimizes real squared error over the sparse rating matrix, clusters of users and thematically similar videos drift together.