ReelSense

An Intelligent Hybrid Movie Recommendation System
Combining Collaborative Filtering + Content-Based Features with Explainability

Features • Tech Stack • Quick Start • How It Works • Metrics • Structure

---

Features

Feature	Description
Hybrid Recommendations	Combines collaborative filtering with content-based methods for superior accuracy
Smart Explainability	Every recommendation comes with a human-readable explanation
Comprehensive Evaluation	Includes Precision@K, Recall@K, diversity, and coverage metrics
Popularity-Based Fallback	Cold-start handling with weighted popularity scoring
Long-Tail Analysis	Visualizes and addresses item popularity distribution
Leave-Last-N Split	Proper temporal train-test splitting for recommender systems

---

Tech Stack

Python

Pandas

NumPy

Scikit-learn

Jupyter

---

Quick Start

Prerequisites

• Python 3.8+
• pip / pip3
• Git

Installation

Step 1: Clone the Repository

git clone https://github.com/sciguy-code/reelsense.git
cd reelsense

---

Step 2: Create Virtual Environment

macOS

python3 -m venv venv
source venv/bin/activate

Linux (Ubuntu/Debian)

# Install venv if not available
sudo apt install python3-venv

# Create and activate
python3 -m venv venv
source venv/bin/activate

Windows (PowerShell)

python -m venv venv
.\venv\Scripts\Activate.ps1

Note: If you get an execution policy error, run:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

Windows (Command Prompt)

python -m venv venv
venv\Scripts\activate.bat

---

Step 3: Install Dependencies

pip install -r requirements.txt

---

Step 4: Launch Jupyter Notebook

jupyter notebook reel_sense.ipynb

Tip: If jupyter is not found, install it with:

pip install jupyter

---

How It Works

Architecture Overview

flowchart TD
    subgraph Pipeline["ReelSense Pipeline"]
        A["Data Loading<br/>(CSV Files)"] --> B["Data Cleaning<br/>& Processing"]
        B --> C["Train/Test<br/>Split"]
        C --> D
        
        subgraph D["Hybrid Recommender Engine"]
            E["Collaborative Filtering<br/>(Cosine Similarity)"] 
            F["Content-Based<br/>(TF-IDF on Genres)"]
        end
        
        D --> G["Explainability Layer<br/>\"Recommended because you liked X\""]
    end
    
    style Pipeline fill:#1a1a2e,stroke:#16213e,color:#fff
    style D fill:#0f3460,stroke:#e94560,color:#fff
    style G fill:#533483,stroke:#e94560,color:#fff

Loading

Algorithm Details

Step 1: Data Preprocessing

• Timestamp Conversion: Unix timestamps to datetime objects
• Year Extraction: Regex-based extraction from movie titles
• Genre Parsing: Pipe-separated genres to lists
• Noise Filtering:
  • Movies with < 5 ratings removed
  • Users with < 20 ratings removed

Step 2: Collaborative Filtering

Uses Item-Item Similarity computed via cosine similarity:

similarity = item_vectors @ item_vectors.T
cf_scores = user_ratings @ similarity

Step 3: Content-Based Filtering

Leverages TF-IDF Vectorization on genre metadata:

tfidf_matrix = TfidfVectorizer().fit_transform(genres)
genre_sim = cosine_similarity(tfidf_matrix)

Step 4: Hybrid Score Blending

final_score = α × CF_score + (1 - α) × Content_score

Where α = 0.7 (adjustable weight parameter)

---

Evaluation Metrics

Model Performance

K Value	Precision@K	Recall@K
K=5	0.63%	3.16%
K=10	0.50%	4.98%
K=20	0.47%	9.30%

Diversity & Coverage

Metric	Value
Catalog Coverage	6.20%
Unique Items Recommended	604

Comparison: Hybrid vs Popularity-Based

Hybrid Model (K=10):      Precision: 0.50%  |  Recall: 4.98%
Popularity-Based (K=10):  Precision: 0.25%  |  Recall: 2.49%

Hybrid Model Improvement: ~2x better performance

---

Project Structure

reelsense/
├── reel_sense.ipynb      # Main Jupyter notebook with full implementation
├── requirements.txt      # Python dependencies
├── README.md             # Project documentation
├── ml-latest-small/      # MovieLens dataset
│   ├── movies.csv        # Movie metadata (9,742 movies)
│   ├── ratings.csv       # User ratings (100,836 ratings)
│   ├── tags.csv          # User-generated tags
│   └── links.csv         # External database links
└── venv/                 # Virtual environment (not tracked)

---

Dataset

This project uses the MovieLens ml-latest-small dataset:

Statistic	Value
Total Ratings	100,836
Total Movies	9,742
Total Users	610
Rating Scale	0.5 - 5.0
Time Range	1996 - 2018

Data Source

F. Maxwell Harper and Joseph A. Konstan. 2015. The MovieLens Datasets: History and Context. ACM Transactions on Interactive Intelligent Systems (TiiS) 5, 4: 19:1–19:19.

---

Sample Recommendations

For User #1 (Action/Sci-Fi Enthusiast):

Movie	Genres	Explanation
Fifth Element, The	Action|Adventure|Comedy|Sci-Fi	You liked 'Star Wars: Episode VI', sharing Action, Sci-Fi, Adventure
Austin Powers	Action|Adventure|Comedy	You liked 'Austin Powers: International Man of Mystery', sharing Action, Comedy, Adventure
Die Hard	Action|Crime|Thriller	You liked 'Shaft', sharing Action, Thriller, Crime
Aliens	Action|Adventure|Horror|Sci-Fi	You liked 'Star Wars: Episode VI', sharing Action, Sci-Fi, Adventure
Hunt for Red October	Action|Adventure|Thriller	You liked 'Dr. No', sharing Action, Thriller, Adventure

---

Future Improvements

• Matrix Factorization (SVD/ALS)
• Deep Learning approaches (Neural Collaborative Filtering)
• Web interface with Flask/FastAPI
• Real-time recommendation updates
• Mobile-responsive frontend

---

Authors

Made with ❤️ for BrainDead 2026 Hackathon

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Made by team CrackHeads for BrainDead 2026 Hackathon.Star this repo if you found it helpful!