Back to Projects
RecSys Challenge 2024 – News Recommendation System Development

RecSys Challenge 2024 – News Recommendation System Development

Recommender Systems
Machine Learning
Large-Scale Data
Algorithm Optimization
News Recommendation

Vienna University of TechnologyMarch 2024 - June 2024

Algorithm Developer

Project Overview

Collaborated in a team of five to implement and improve three algorithms, including NRMS, Standard VAE, and SAR, for large-scale news recommendation. Focused on addressing beyond-accuracy metrics such as diversity, serendipity, and novelty to align with the ACM RecSys 2024 objectives.

Challenges

  • Working with the large-scale Ekstra Bladet News Recommendation Dataset (600M impressions)
  • Improving baseline and advanced algorithms to meet beyond-accuracy objectives like diversity and serendipity
  • Optimizing training processes and addressing computational constraints using smaller datasets
  • Evaluating and comparing the performance of algorithms under strict challenge criteria

Key Achievements

  • Successfully implemented and improved three algorithms: NRMS, Standard VAE, and SAR
  • Enhanced SAR algorithm, achieving notable improvements in ranking metrics such as NDCG and precision
  • Generated reproducible results using advanced preprocessing and data augmentation techniques
  • Submitted competitive predictions to the RecSys Challenge leaderboard

Technologies Used

NRMS
Standard VAE
SAR
Python
Machine Learning
Recommendation Metrics