Movie recommendation system with Python. Implements content-based filtering (TF-IDF + cosine similarity), collaborative filtering with matrix factorization (TruncatedSVD), and a hybrid approach. Evaluates with Precision@K, Recall@K, and NDCG. Includes rating distribution plots, top movies, and sample recommendations.

• Synthetic ratings dataset with users, movies, and multi-label genres
• Content-based: TF‑IDF on genres + item–item cosine similarity
• Collaborative: user–item matrix factorization (TruncatedSVD)
• Hybrid: weighted combination of content & collaborative scores
• Evaluation: Precision@K, Recall@K, NDCG@K (per-user & macro)
• Visuals: rating distribution, top movies, sample recommendations
• Outputs: metrics JSON, recommendation files, and charts

movie-recommendation-system/
├─ README.md
├─ LICENSE
├─ requirements.txt
├─ data/
│  └─ generate_ratings.py
├─ src/
│  ├─ build_recommender.py
│  └─ utils.py
└─ outputs/
   └─ figures & reports (auto-created)

python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate
pip install -r requirements.txt

python data/generate_ratings.py --users 800 --movies 1200 --seed 42 --outdir data
# Produces: data/ratings.csv, data/movies.csv

python src/build_recommender.py --ratings data/ratings.csv --movies data/movies.csv --outdir outputs --k 10 --alpha 0.6 --seed 42

Outputs

• outputs/metrics.json – Precision@K, Recall@K, NDCG@K (content, collab, hybrid)
• outputs/recs_user_*.txt – sample recommendations for a few users
• outputs/ratings_hist.png
• outputs/top_movies.png

➡️ On this sparse synthetic dataset, content-based slightly outperforms collaborative and hybrid.

User 1

Movie 959    41.08
Movie 541    36.28
Movie 952    34.83
Movie 95     34.81
Movie 40     33.30

User 2

Movie 915    47.02
Movie 122    38.27
Movie 77     38.18
Movie 1097   37.65
Movie 952    37.04

User 3

Movie 800    52.00
Movie 410    45.77
Movie 9      45.59
Movie 738    41.34
Movie 713    41.00

• Positive items are defined as ratings ≥ 4 for evaluation.
• The hybrid weight alpha controls emphasis on collaborative (α→1) vs. content (α→0).