About this repository

This repository is created for store, record POSTECH DL Project - Sentimeter(Group 20)

To revisit the experiments

0. Baseline

• data in ./Bert-Tutorial/data/
• experiment in ./Bert-Tutorial/Bert-custom_linear.ipynb

1. Data Augmentation

• data in ./data/augmented/
• augmentation in ./data_augmentation
• experiment with augmented data in ./Bert-Tutorial/Bert-custom-augment.ipynb

2. Embedding layer

• data in
• trial 1 in ./BertCNNsenti_analysis/BertCharCNNmodel.ipynb
• trial 2 in
• trial 3 in ./Bert-Tutorial/Bert-TEST.ipynb

3. Classifier

• data in ./Bert-Tutorial/data/
• trial 1 in
• trial 2 in
• trial 3 in ./Bert-Tutorial/Bert-TEST.ipynb

4. Ensemble

• data in
• experiment in ./BertCNNsenti_analysis/BertCharCNNmodel.ipynb

Reference papers we read

• Fine-grained Sentiment Classification using BERT (AITB'19)

• Convolutional Neural Networks with Recurrent Neural Filters (EMNLP'18)

• Deep contextualized word representations (NAAAL'18)

• Convolutional Neural Networks for Sentence Classification (EMNLP'14)

• A multi-sentiment-resource enhanced attention network for sentiment classification (arXiv preprint arXiv:2018)

• Self-Explaining Structures Improve NLP Models (arXiv preprint arXiv:2020)

• Bag of Tricks for Efficient Text Classification (arXiv preprint arXiv:2016)

• Convolutional Neural Networks for Sentence Classification (EMNLP'14)

• Recurrent Neural Network for Text Classification with Multi-Task Learning (IJCAI'16)

• Feed-Forward Networks with Attention Can Solve Some Long-Term Memory Problems (arXiv preprint arXiv:2015)

• Recurrent Convolutional Neural Networks for Text Classification (AAAI'15)