Movie-Recommender-System

This repository contains the implementation of a deep learning-based recommender system using Neural Collaborative Filtering (NCF). The project aims to accurately predict user-item interactions from implicit feedback data from the The Movie Small dataset.

Key Technical Features

• Model Architecture: Implements the NeuMF structure in PyTorch, combining the efficiency of Generalized Matrix Factorization (GMF) with the expressive power of a Multi-Layer Perceptron (MLP) to capture complex non-linear user-item interaction functions.
• Implicit Feedback: The problem is framed as a binary classification task, optimizing with Binary Cross-Entropy Loss (BCELoss).
• Data Integrity: Includes a custom, optimized negative sampling function that strictly adheres to the Leave-One-Out (LOO) protocol. This prevents data leakage by ensuring that no training interaction is accidentally used as a negative sample in the test set.
• Cold-Start Mitigation: Implements Item Frequency Filtering to remove cold-start movies (items with low interaction counts, e.g., $<3$) from the evaluation pool, ensuring the performance metrics (NDCG/Hit Rate) reflect the model's true capability on learnable items.

Repo Structures

1. dataset_preparation.ipynb : This is the part 1 where transform the raw data into training and testing dataset.
2. movie_NCF.ipynb : This is the part 2 where build the NCF model, training, fine-tune, and evaluation.
3. train.csv : dataset used in the model training.
4. test.csv : dataset used in the evaluation.