Ekstra Bladet new reranker

This repository contains the code and the pre-trained models for the DArgk team submission to the ACM RecSys Challenge 2024.

Proprocessing

The notebook 01-DataPreprocess.ipynb is intended for preprocessing the articles, behaviors and behaviors part of the dataset.

The notebook 02-Img-embs.ipynb encodes the Resnet embedding of the images into a more 128-dimension vector using an autoencoder.

Generating the predictions

This repository contains all the required files, but the dataset, for generating the rankings as submitted to the challenge. Steps for reproducing the results:

1. Recreate the conda environment using the environment.yml file.

conda env create -f environment.yml

2. Place the dataset files as provided in the folder dataset.
3. Generate the predictions.

python inferv1_img_bce_val.py

4. To format the output of the predictions.

python to_zip_format.py --exp v1_img_bce_val_epoch_9

Training the model

The model was trained for 8 epochs in the training dataset and finetuned for 1 each on validation.

For executing one epoch of training in the training dataset:

python train_1_img_bce_epoch.py

For fine-tuning one epoch using the validation dataset:

python train_1_img_bce_val_epoch.py

Cite

@inproceedings{10.1145/3687151.3687161,
author = {Rodriguez, Juan Manuel and Tommasel, Antonela},
title = {Leveraging User History with Transformers for News Clicking: The DArgk Approach},
year = {2024},
isbn = {9798400711275},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
doi = {10.1145/3687151.3687161},
pages = {48–52},
numpages = {5},
location = {Bari, Italy},
series = {RecSysChallenge '24}
}

The paper is open access and can be found at ACM Digital Library

Contact info:

• Antonela Tommasel (antonela.tommasel@isistan.unicen.edu.ar)
• Juan Manuel Rodriguez (jmro@cs.aau.dk)