🎬 Hybrid Movie Recommendation System

A hybrid recommender system combining User-Based and Item-Based Collaborative Filtering on the MovieLens 20M dataset to generate personalized movie recommendations.

Problem Statement

Given a user's movie rating history, recommend new movies they are likely to enjoy using two complementary approaches:

• User-Based CF — Find similar users by taste and recommend what they liked
• Item-Based CF — Find movies similar to what the user already rated highly

Dataset

• Source: MovieLens 20M Dataset
• Scale: 20 million ratings from 138,493 users across 27,278 movies
• After filtering: Movies with 1,000+ ratings retained (3,134 movies)

Algorithm

User-Based Collaborative Filtering

1. Build user-movie rating matrix (138K × 3K)
2. Identify users who watched ≥60% of the same movies as the target user
3. Compute Pearson correlation between target user and candidates
4. Filter to users with correlation ≥ 0.65
5. Calculate weighted ratings (correlation × rating)
6. Rank movies by weighted average score → Top 5 recommendations

Item-Based Collaborative Filtering

1. Find the target user's highest-rated recent movie
2. Compute pairwise Pearson correlation between that movie and all others
3. Rank by correlation strength → Top 5 recommendations

Hybrid Output

Combines both approaches: 5 user-based + 5 item-based = 10 recommendations

Example Output

User-Based Recommendations:

Movie	Weighted Score
The Shawshank Redemption	4.52
Pulp Fiction	4.38
Forrest Gump	4.31
The Matrix	4.27
Fight Club	4.15

Actual results vary by target user.

Tech Stack

• Python 3.8+ — Core language
• Pandas — Data manipulation, pivot tables, correlation computation
• NumPy — Numerical operations

Getting Started

git clone https://github.com/eboekenh/Movie_Recommendation_Algorithm.git
cd Movie_Recommendation_Algorithm
pip install -r requirements.txt

Download the MovieLens 20M dataset and place movies.csv and ratings.csv in the project directory.

python Hybrid_movie_recommendation.py

Key Design Decisions

• Minimum 1,000 ratings per movie — Ensures correlation calculations are statistically meaningful
• 60% overlap threshold — Users must share substantial viewing history for reliable similarity
• Pearson correlation ≥ 0.65 — Only strongly similar users influence recommendations
• Weighted ratings — Higher-correlated users have more influence on recommendations

License

MIT