Course contentsShow
← Machine Learning and Deep Learning
Lesson 2346 of 3,53851. Recommender SystemsFree lesson

Weighted User Profiles

Incorporating recency, frequency, and explicit ratings to weight features differently in user profiles.

Weighted User Profiles

What you'll learn: How to build smarter user profiles by giving more importance to recent interactions, frequent behaviors, and explicit ratings rather than treating all user actions equally.

The Core Idea

Not all user interactions are created equal. If you watched a movie yesterday, that's more relevant than one you watched three years ago. If you've rated thriller movies highly five times, that signal is stronger than a single click on a comedy. Weighted user profiles adjust the importance of different features in a user's profile based on three key factors:

  • Recency: How recent was the interaction? Recent preferences matter more.
  • Frequency: How often does the user engage with this feature? Repeated behavior signals strong preference.
  • Explicit ratings: Did the user provide a direct rating (like 5 stars)? Explicit feedback trumps implicit clicks.

Building Weighted Profiles

Think of a user profile as a shopping basket, but each item has a different "weight" sticker. A thriller movie watched yesterday might have weight 2.5, while a comedy from two years ago gets weight 0.3. When computing similarity between the user's profile and candidate items, these weights multiply the feature values.

For example:

  • A user watched three sci-fi movies (frequency = 3) → higher weight
  • They rated action movies 4.5/5 (explicit rating) → high weight
  • They clicked on a romance movie once, two years ago → low weight

When you construct the user's feature vector, you multiply TF-IDF scores (or other feature representations) by these combined weights: weighted_score = base_score × recency_weight × frequency_weight × rating_weight.

Why This Matters

Without weighting, a single old interaction can pollute recommendations. Weighted profiles ensure your system adapts to evolving tastes and emphasizes the signals that truly matter.

Key Takeaway: Weighted user profiles improve content-based filtering by incorporating recency decay, frequency amplification, and explicit rating emphasis, making recommendations more accurate and responsive to current user preferences.