Multi-task Recommender System Using Tensorflow Recommenders

(Under Development)

Click here to Open NoteBook to see Approach Code

This project implements a multi-task recommender system using TensorFlow Recommenders (TFRS). It's based on the TensorFlow Recommenders example for building a multi-objective recommender for Movielens, using both implicit (movie watches) and explicit signals (ratings).

Project Structure

multitask_recommender/
├── config.yaml       # Configuration file for hyperparameters
├── data.py         # Script for loading and preprocessing the Movielens dataset
├── model.py        # Script defining the multi-task recommender model
├── train.py        # Script for training and evaluating the model
├── requirements.txt # List of project dependencies
└── README.md       # This file

Configuration

The config.yaml file contains the following hyperparameters:

• embedding_dimension: Dimension of the embedding vectors for users and movies.
• rating_weight: Weight for the rating prediction loss.
• retrieval_weight: Weight for the retrieval task loss.
• learning_rate: Learning rate for the optimizer.
• epochs: Number of training epochs.
• batch_size: Batch size for training.

Dependencies

The project dependencies are listed in requirements.txt. To install them, run:

pip install -r requirements.txt

Running the project

To train and evaluate the model, run the train.py script:

python train.py

This script will:

1. Load the Movielens dataset.
2. Build the multi-task recommender model.
3. Train the model using the training data.
4. Evaluate the model on the test data, reporting the retrieval top-100 accuracy and the ranking Root Mean Squared Error (RMSE).

Model Description

The model is a multi-task model that jointly optimizes for two tasks:

1. Rating Prediction: Predicts the explicit ratings given by users to movies.
2. Retrieval: Predicts which movies a user will interact with (implicit feedback).

The model consists of:

• User and movie embedding layers.
• A rating prediction network that takes user and movie embeddings as input.
• Two task-specific loss functions: Mean Squared Error for rating prediction and a retrieval loss.
• Weighted combination of the two losses to train the model jointly.

Getting Started

1. Clone the repository.
2. Navigate to the multitask_recommender directory.
3. Install the dependencies: pip install -r requirements.txt.
4. Run the training script: python train.py.

Further Improvements

• Experiment with different loss weights to balance the importance of the two tasks.
• Explore different model architectures for the rating prediction network.
• Incorporate additional user and movie features.
• Evaluate the model with different metrics.