BERT-XLNet-Irony

Evalution of Fine-tuning BERT and XLNet on Irony Detecion in English Tweets

For the task description, check two things:

1. explanation on CodaLab
2. SemEval-2018 Task 3: Irony Detection in English Tweets

This project provides the source code of the paper "Evaluation of BERT and XLNet Models on Irony Detection in English Tweets" which was presented on DEIM2020 and got a presentation award. DEIM2020 is a japanese workshop. The fullname of it is Proceedings of Forum on Data Engineering and Information Management.

1. Environment

python3.6
jupyterlab==1.2.4
matplotlib==3.1.1
nltk==3.4.5
numpy==1.17.4
pandas==0.25.3
scikit-learn==0.22
scipy==1.3.2
torch==1.1.0s
torchvision==0.3.0
transformers==2.3.0

2. Data Preprocessing

After setting the environment, we should process the tweets for training. Just run the data_preprocess.ipynb to get the normalized_sents.pickle

3. Word Embedding method

Check the embedding_method for details.

3. Fine-tuning method

Check the finetuning_method for details.