Movies-Recommendation-System-Project

A movie recommendation system built using the MovieLens dataset that suggests top movies to users based on their ratings using collaborative filtering and

Project Overview

This project builds a movie recommendation system using both:

• Collaborative Filtering (SVD)
• Content-Based Filtering (TF-IDF)

These approaches are combined into a hybrid model to improve recommendation quality and address real-world challenges such as the cold-start problem.

---

Business Problem

Streaming platforms need to recommend relevant content to users. However:

• New users have little interaction data
• New movies may not have ratings
• Single-model systems are limited

This project develops a hybrid recommendation system that balances:

• Personalization (user preferences)
• Content similarity (movie features)

---

Data Understanding

The dataset includes:

• User ratings
• Movie titles
• Genres
• Tags

Key Steps:

• Data loading and inspection
• Handling missing values
• Removing duplicates
• Exploratory analysis (rating distributions, sparsity)

---

Data Preparation

• Merged ratings and movie metadata
• Filtered low-activity users and movies
• Created combined text features (genres + tags)

---

Modeling

1. Baseline Model

• Simple collaborative filtering benchmark

2. SVD Model

• Matrix factorization using Singular Value Decomposition
• Captures latent relationships between users and movies

3. Tuned SVD Model

• Hyperparameter tuning applied
• Improved performance across evaluation metrics

---

Model Evaluation

Models were evaluated using:

• RMSE (Root Mean Squared Error)
• MAE (Mean Absolute Error)
• Precision

Tuned SVD achieved the best predictive performance

---

Content-Based Filtering (TF-IDF)

• Applied TF-IDF to genres and tags
• Converted text into numerical vectors
• Used cosine similarity to measure movie similarity

Purpose:

• Recommend similar movies
• Handle the cold-start problem

---

Hybrid Recommendation System

The hybrid model combines:

• Predicted ratings (SVD) → personalization
• Similarity scores (TF-IDF) → content relevance

Hybrid Model Insights

• Tested different alpha values
• Observed how recommendations change:
  • Lower α → more similarity-based
  • Higher α → more personalized

---

Recommendation Functions

Collaborative Filtering

• Recommends unseen movies using tuned SVD

Hybrid Model

• Combines SVD predictions with TF-IDF similarity
• Produces more balanced recommendations

---

Results & Insights

• Tuned SVD improved prediction accuracy
• TF-IDF enabled content-based similarity
• Hybrid model provided a more flexible and robust system

---

Limitations

• TF-IDF depends on quality of text data
• Hybrid model requires tuning (alpha selection)
• Evaluation differs between ranking and prediction tasks

---

Future Improvements

• Use advanced NLP techniques (e.g., embeddings)
• Improve feature engineering
• Explore deep learning recommendation models
• Deploy as a web application

---

Tableau

Conclusion

This project demonstrates that:

• Model tuning significantly improves performance
• Collaborative filtering alone is not sufficient
• Content-based methods help solve cold-start problems
• Hybrid models provide a more practical real-world solution

Technologies Used

• Python
• Pandas, NumPy
• Scikit-learn
• Surprise (SVD)
• TF-IDF (NLP)
• Tableau

Setup Instructions

1. Clone the repository git clone https://github.com/Marian-amo/Movies-Recommendation-System-Project

2. Create a virtual environment (recommended) python -m venv venv

3. Install dependencies pip install -r requirements.txt

4. Download Dataset https://grouplens.org/datasets/movielens/

5. Run Jupyter Notebook jupyter notebook Open:notebook.ipynb