Cross-lingual Product Matching using Transformers

Can matching knowledge be transferred between languages?

• Can we augment performance on languages with small amounts of training data by adding data from more readily available languages (English)? (E.g. Fine-tune model for the task using english data / Further fine-tune using few examples of target language (few-shot learning))
• Can we achieve good performance without any data from the target language? (Zero-shot performance)
• If it works well, how can we explain it?

Contributors

These people have contributed to this repository:

• @Progandi
• @bebing
• @daniels9
• @fniesel
• @JakobGutmann

Setup

How to use this Repository

Before installing, make sure to have Microsoft Build Tools for C++ installed for fastText

cd path/to/repo
conda env create -f environment.yml
conda activate team_project

Data Our datasets can be requested via mail at ralph@informatik.uni-mannheim.de, but are available for research purposes only.