Deep Matrix Factorization and Interaction Functions
What you'll learn: How neural networks generalize classic matrix factorization by learning complex interaction patterns between user and item embeddings.
Beyond Linear Dot Products
Traditional matrix factorization predicts ratings using a simple dot product between user and item embeddings: rating = user_embedding · item_embedding. This assumes a linear relationship between features. But what if user preferences are more complex?
Deep matrix factorization replaces this rigid dot product with learnable interaction functions—neural network layers that can capture non-linear patterns in how users relate to items.
The Evolution of Interaction Modeling
Think of it like cooking: traditional matrix factorization is mixing two ingredients and tasting them (dot product). Deep matrix factorization is like having a chef (neural network) who can combine, transform, and cook those ingredients in sophisticated ways before serving the final dish (prediction).
Common Interaction Functions
- Concatenation + MLP: Concatenate user and item embeddings, then pass through fully connected layers that learn complex feature interactions
- Element-wise Product + MLP: Multiply embeddings element-wise first (like classic MF), then transform through neural layers for added expressiveness
- Hybrid Approaches: Combine dot product (generalized matrix factorization) with MLPs in parallel, capturing both linear and non-linear patterns
Why This Matters
The interaction function is where the "magic" happens—it determines how much modeling power your recommender has. A dot product gives you classic collaborative filtering. A deep network can learn that "users who like action movies AND rate highly also prefer specific directors," capturing multi-way interactions impossible with linear methods.
Key Takeaway: Deep matrix factorization generalizes collaborative filtering by replacing fixed dot products with learnable neural interaction functions, enabling the model to discover complex, non-linear relationships between users and items.