Item Feature Representation
What you'll learn: How to encode items as numerical feature vectors that capture their attributes for use in recommender systems.
The Core Idea
In content-based filtering, we can't just feed raw items (movies, products, articles) into our models. We need to translate each item into a feature vector—a list of numbers that mathematically represents what that item is. Think of it like creating a standardized "profile card" for every item, where each field is a number our model can process.
From Attributes to Vectors
Items come with various types of information:
- Tags and categories: "action", "sci-fi", "comedy"
- Text descriptions: movie plots, product reviews
- Metadata: release year, price range, brand
- Multi-valued attributes: actors, directors, ingredients
The challenge is converting these diverse attributes into a consistent numerical format.
Representation Strategies
One-hot encoding works for categorical features: if a movie's genre is "sci-fi" from three possible genres, represent it as [0, 1, 0].
TF-IDF vectors capture textual descriptions, turning words into weighted importance scores.
Embeddings can map high-dimensional sparse features (like thousands of possible tags) into dense, lower-dimensional vectors that capture semantic similarity.
Concatenation combines different feature types: stack genre vectors, year (normalized), and average rating into one comprehensive vector like [0, 1, 0, 0.75, 4.2].
Why This Matters
Once items are vectors, you can compute similarity (via cosine distance or dot products) and feed them into neural networks. A user who liked items with vector [0.8, 0.2, 0.5] might enjoy items close to that pattern in vector space.
Key Takeaway: Item feature representation transforms heterogeneous item attributes into standardized numerical vectors, enabling mathematical comparison and learning in content-based recommender systems.