English
- Stanford Sentiment Treebank
- The corpus is based on the dataset introduced by Pang and Lee (2005) and consists of 11,855 single sentences extracted from moview reviews.
- Each phrase is labeled as either negative, somewhat negative, neutral, somewhat positive or positive. (SST-5 or SST fine-grained)
- train: 8,544
- dev: 1,101
- test: 2,210
- https://paperswithcode.com/dataset/sst
Korean
- Naver Sentiment Movie Corpus
- The corpus is a movie review dataset in the Korean language. Reviews were scraped from Naver Movies.
- 200K reviews in total
- ratings_train.txt: 150K reviews for training
- ratings_test.txt: 50K reviews held out for testing
- https://github.com/e9t/nsmc/
English
- GloVe
- BERT (work in progress)
Korean
- Character Embedding
- BERT (work in progress)
English
GloVe + CNN
- Validation accuracy
- Test accuracy
GloVe + BiLSTM
- Validation accuracy
- Test accuracy
Korean
Character Embedding + CNN
- Test accuracy
Character Embedding + BiLSTM
- Test accuracy
