Content-Based TF-IDF

Content-Based TF-IDF builds item profiles from text descriptions or metadata using TF-IDF weighting, then recommends items with the highest cosine similarity to items the user has previously liked.

When to use:

• Cold start for new users — no interaction history needed, only item content
• Catalog-heavy recommendations (articles, products with descriptions)
• When item metadata is rich and meaningful for preference modeling

Input: User interaction history + item text/metadata columns Output: Ranked list of content-similar items per user

Model Settings (set during training, used at inference)

Text Columns (set during training) Which item columns are used for TF-IDF feature extraction.

Max Features (default: 10000) Maximum vocabulary size for TF-IDF vectorization.

N Recommendations (default: 10) Number of top-N items to return per query.

Inference Settings

No dedicated inference-time settings. Items are ranked by TF-IDF cosine similarity to the user's profile.