Popularity measures how broadly items are favored, viewed, or selected. In recommender systems, it provides a non-personalized signal and a practical starting point before personalization. section 2.0 section 4.4.4
Use popularity in three closely related ways:
- Recommend the most popular catalog items to every user as a simple baseline. section 4.4.4
- Add popularity to candidate scores alongside similarity, recency, genre preference, or neural scores. section 2.2.4
- Start with popular items and personalize their selection or ordering for a user. section 1.2.1
Normalized movie popularity
Count ratings for each movie, divide each count by the maximum count, and assign unseen items a score of 0.0. In one catalog, Forrest Gump scores 1.000 from 81,491 ratings, while Pulp Fiction scores 0.977 from 79,672 ratings. section 2.2.4
Listing 2.7 Computing popularity scores
movie_counts = ratings['movieId'].value_counts() #A
max_count = movie_counts.max() #B
popularity_scores = {
int(mid): float(count / max_count)
for mid, count in movie_counts.items()
}
Shows the compact computation that turns item interaction counts into normalized popularity scores. section 2.2.4
Popularity is difficult to beat because a small number of items account for many interactions. Compare it before trusting a more complex model: outperforming it suggests the model learned user-specific patterns, while losing to it may expose implementation or data-pipeline problems. section 4.5.1 section 4.4.4

