Learning-to-rank algorithms

This repository contains example implementations of deep learning and gradient boosting based algorithms for the task of learning to rank a set of candidates to find for a user an approximate rating of a movie based on what the user has rated before.

Models

The following models are tested within this repository:

• LGBMRanker (Website)
• DeepFM (Paper)
• DCNv2 (Paper)

Dataset

The repository contains a dataset loader for the Movie-lens 100K dataset accesible here. The contents can be extracted into the data folder and the code should work without any changes.

Examples

To train the model with DeepFM, run:

uv run -m src.main --model deepfm

and with DCNv2:

uv run -m src.main --model dcnv2

The model parametrs can be configured by changing the files within the config folder.