11-785ActiveLearningProject

This is the repository of 11-785 course project:

Incorporating Active Learning with Self-Training based Abusive Language Detection

Group Members (Andrew ID):

• Chuiyu Wang (chuiyuw)
• Yimeng Gu (yimengg)
• Haoyue Zhao (haoyuez)
• Xiang Liu (xliu4)

1. Datasets

1. English Datasets:

https://drive.google.com/drive/folders/1NjCtBvOKRjCiVbmEWKmqhHiweW5NdhtH?usp=sharing

2. Turkish Datasets:

https://drive.google.com/drive/folders/1HK_NeTg_PN7iRwl73O9RD8xUVBUhUjda?usp=sharing

2. Code Preparation

• Download the English datasets to the ./EnglishDataset/ directory
• Download the Turkish datasets to the ./TurkishDataset/ directory
• Clone the huggingface xlm-roberta repository
• Check if the file of pretrained model parameters is around 1GB, if not, use wget to re-download it (already scripted in a code cell)

3. Code Running

• Run the code cell step by step
• If you have already cloned the huggingface xlm-roberta repository, skip the relevant code cell
• The code contans 2 stages:

1. Pretrain the model on English dataset

2. Finetune the model on Turkish dataset