Multitask Recommender System

This repository implements a Multi-Task Neural Network for Recommender Systems using PyTorch. The model is designed to handle both implicit feedback (via Matrix Factorization) and explicit feedback (via Regression) simultaneously. It utilizes the MovieLens dataset for training and evaluation.

Features

• Multi-Task Learning: Combines Matrix Factorization (BPR loss) and Regression (MSE loss) in a single model.
• Flexible Architecture: Supports shared or separate embeddings for the factorization and regression tasks.
• MovieLens Integration: Automatically downloads and processes MovieLens datasets (default: 100K).
• Evaluation Metrics: Tracks Mean Reciprocal Rank (MRR) and Mean Squared Error (MSE).
• TensorBoard Support: Logs training metrics for visualization.

Setup

Ensure you have Python installed. You can install the required dependencies using pip:

python -m pip install -r requirements.txt

or by creating a Virtual Envirmoment and installing the dependencies by running the shell script (Recommended).

bash setup.sh

Usage

To train the model, run the src/main.py script. You can configure the training process using command-line arguments.

Basic Example

python src/main.py --epochs 10

Custom Configuration

python src/main.py \
    --logdir logs/experiment1 \
    --test_fraction 0.2 \
    --shared_embeddings \
    --factorization_weight 0.5 \
    --regression_weight 0.5 \
    --epochs 20

Arguments

Argument	Type	Default	Description
--test_fraction	float	0.05	Fraction of the dataset to use for testing.
--epochs	int	1	Number of training epochs.
--factorization_weight	float	0.995	Weight for the factorization task loss.
--regression_weight	float	0.005	Weight for the regression task loss.
--shared_embeddings	flag	True	Use shared embeddings for both tasks.
--no_shared_embeddings	flag	False	Use separate embeddings for factorization and regression.
--logdir	str	run/...	Directory for TensorBoard logs.
--gpu	bool	False	Enable GPU training.

Project Structure

• src/main.py: Entry point for training the model.
• src/multitask.py: Implements the MultitaskModel class handling the training loop.
• src/models.py: Defines the MultiTaskNet neural network architecture.
• src/dataset.py: Utilities for downloading and loading MovieLens data.
• src/losses.py: Custom loss functions (BPR, Regression).
• src/evaluation.py: Evaluation metrics (MRR, MSE).
• src/utils.py: Helper functions for tensor manipulation and batching.