This repository contains the code for replicating the experiments around the use of Transformer-based language models for cross-lingual product matching which have been conducted by a student team project in Spring 2021. More information about the project and its results is found on the project's webpage Cross-lingual Product Matching using Transformers.
The project seeked to explore whether learned matching knowledge in the product matching domain can be transferred between languages. More specifically, we investigate whether high-resource languages, such as English, can be used to augment performance in scenarios where either no (zero-shot) or few (few-shot) examples are available in the target language. Towards that, we also study the effect of cross-lingual pre-training with models like mBERT or XLM-R and compare them to monolingual Transformer-architectures and simple baseline models.
These people have contributed to this repository:
Our datasets can be requested via mail at ralph@informatik.uni-mannheim.de, but are available for research purposes only. The datasets have to be included in a separate subfolder 'datasets' in the project folder for the experiments.
How to use this Repository
Before installing, make sure to have Microsoft Build Tools for C++ installed for py_entitymatching
Individual experiments can be configured using the .json settings files.
The settings_template.json provides an overview over the possible settings for the experiments.
Some settings are only available in the multi-class setup but not in the pair-wise case, and vice versa.
To run a experiment, make sure to provide the path of the individual .json settings file
as input agrument. For instance, to run settings_baseline_multi.json, include the argument:
--i path_to_file\settings_baseline_multi.json