Recommender Systems - Vienna University of Technology
Final Report in PDF
WeKnowWhatYouWant presents our participation in the ACM RecSys 2024 Challenge using the Ekstra Bladet News Recommendation Dataset (EB-NeRD). This dataset, gathered from the Danish news site Ekstra Bladet, contains user interaction logs over six weeks and includes comprehensive article metadata.
Our project employs two state-of-the-art recommendation algorithms:
- Neural Recommendation with Personalized Attention (NPA)
- Neural Recommendation with Long- and Short-term User Representations (LSTUR)
We aim to enhance these models to predict user preferences effectively.
To ensure the quality and relevance of our solutions, we evaluate the models using stringent metrics:
- AUC (Area Under Curve)
- MRR (Mean Reciprocal Rank)
- nDCG@K (Normalized Discounted Cumulative Gain)
- Benchmark results will be submitted to the RecSys Challenge leaderboard.
- Comprehensive documentation and reproducibility are ensured through detailed coding and reporting practices.
This project involves downloading datasets and utilizing a variety of machine learning and data processing libraries to work with these datasets. The project is split into multiple components, including downloading files, processing data, and building models.
This script is designed to download a set of files from specified URLs and extract them if they are zip files. It uses the requests library to download the files and tqdm for progress tracking.
This Jupyter Notebook implements NPA algorithm includes code that leverages several machine learning libraries such as transformers, tensorflow, and polars for processing and analyzing datasets. The notebook contains cells to tokenize data, manage paths, and perform various utilities specific to the ebrec package.
The NPA algorithm, or Non-negative Principal Component Analysis, is a method used in data analysis to simplify large datasets while ensuring that the simplified data remains meaningful and easy to interpret. Here's a simple breakdown:
Purpose: The main goal is to reduce the complexity of a large dataset by finding new, smaller sets of data (called components) that still capture the important information from the original data.
Non-negative: The algorithm ensures that all the values in these new components are non-negative, meaning they are zero or positive. This is important in fields like image processing or biology, where negative values don't make sense.
Principal Components: These are the new sets of data that the algorithm finds. They are like summaries of the original data, showing the main patterns or features.
Process: The algorithm looks at the original data, identifies the main features, and creates new components that are easier to work with but still represent the original data well.
This file lists the dependencies required for the project. It includes specific versions of libraries to ensure compatibility.