This technique attempts to explain users' ratings of inventory items (for example, products on Amazon) by inferring a secondary set of latent factors which are inferred from ratings. The power comes from the fact that you do not need to know the factors ahead of time (similar to PCA techniques), but they are simply inferred from the ratings themselves. We derive the latent factors using matrix factorization techniques which are popular due to the extreme scalability, accuracy of predictions, and flexibility (they allow for bias and the temporal nature of the user and inventory).
- Singular Value Decomposition (SVD): SVD has been available in Spark from the early days, but we recommend not to use it as a core ...