Loreley99 / Human-in-the-loop-TUM

This repository based on trec dataset implement different augmentation method including sentence level as well as embedding level and user can simply pick the data they want to augment and check the model performence.

Human-in-the-loop-TUM

This repository based on the TREC dataset implement different augmentation method including the language level as well as the embedding level and user can simply pick the data they want to augment and check the model performance. In addition, this is the project outcome of TUM's XAL practice course. The project report can be found at XAI_Report.

Usage

• Easy data augmentation methods are implemented in eda.py and eda.ipynb.
• Back translation augmentation method is implemented in back_translation.py and back_translation.ipynb, where the German-English and Russian-English back translations are used.
• The embedding level data augmentation method is implemented in augmentation_all_methods.ipynb. Here we also encapsulate the user's discretion to choose how the three methods are combined and the process for augmenting data or data groups.

Data

The data folder contains the augmentation data generated by the three augmentation methods. This data can also be generated by running the corresponding .ipynb file.

Reference

The implementation of the eda and back translation methods is referenced from DoubleMix.

Contact

If there are any questions, welcome to contact us by [e-mail].