Movie Recommendation Systems

This repository contains two movie recommendation systems developed using the MovieLens 100K dataset. It includes both traditional user-based collaborative filtering and a deep learning-based neural collaborative filtering model.

---

Project Structure

File	Description
rs1.py	User-Based Collaborative Filtering with cosine similarity
rs2.py	Neural Collaborative Filtering (NCF) using PyTorch
run_eval.py	Evaluation script computing RMSE and novelty for both systems
eda.ipynb	Jupyter Notebook for Exploratory Data Analysis (EDA)

---

Setup Instructions

1. Download the Dataset

Download the MovieLens 100K dataset and place the extracted ml-100k folder in the project root:

project/
├── ml-100k/
│   ├── u.data
│   ├── u.item
│   ├── u.user
│   ├── u.genre
├── rs1.py
├── rs2.py
├── run_eval.py
├── eda.ipynb
└── README.md

---

How to Run

When both models are first run, it takes some time for them to train. After that, they are stored as a pickle model, which can later be used without training.

RS1: User-Based Collaborative Filtering

python rs1.py

• Computes cosine similarity between users
• Uses top-K nearest neighbors for prediction
• Prompts user input and returns top 10 movie recommendations

RS2: Neural Collaborative Filtering (NCF)

python rs2.py

• Trains a PyTorch model with user/movie embeddings and MLP layers
• Saves best model based on RMSE
• Recommends top 10 movies based on learned user preferences

---

Evaluation

Run the following to compare the two systems:

python run_eval.py --num_users 10

Metrics evaluated:

• RMSE: Accuracy of predicted ratings
• Novelty: Encourages recommendations beyond the most popular items

---

Models Used

RS1

• Collaborative filtering with cosine similarity
• Dynamic K-selection for neighbors

RS2

• Neural network with embedding layers for users and items
• Hidden layers with dropout, batch normalization, and LeakyReLU

---

EDA

The eda.ipynb notebook includes:

• Distribution of ratings
• User/movie interaction statistics
• Insights on data sparsity

---

Demo

Once a model is trained, the system interactively asks for a user_id:

Enter User ID to get recommendations:

And returns:

Top 10 recommended movies for User X:
1. Movie A
2. Movie B
...

---

Requirements

• Python 3.8+
• pandas, numpy, matplotlib, scikit-learn
• torch, torchvision (for RS2)

Install with:

pip install -r requirements.txt

---