Implementation of some aspect-based sentiment analysis models; based on car dataset/ laptop dataset; 3 classes
-
ATAE-LSTM(Attention-based LSTM with Aspect Embedding)
Attention-based LSTM for Aspect-level Sentiment Classification -
TSA(A Siamese Bidirectional LSTM with context-aware attention) DataStories at SemEval-2017 Task 4: Deep LSTM with Attention for Message-level and Topic-based Sentiment Analysis
| item | value |
|---|---|
| training set | 12813 |
| valid/dev set | 1602 |
| test set | 1602 |
| char_vocab | 2378 |
| all_char_vocab | 2379 |
| aspect | 20 |
| aspect_text_char_vocab | 69 |
| char_max_len | 127 |
| < char_len = 0.991 | 110 |
| aspect_text_char_max_len | 19 |
| < aspect_text_char_len = 0.978 | 18 |
- random
| model | acc(on test) | acc(on dev) | macro-f1(test) | macro-f1(dev) |
|---|---|---|---|---|
| atae_lstm | 0.6536 | 0.6685 | 0.5952 | 0.6143 |
| tsa | 0.6654 | 0.6816 | 0.6194 | 0.6381 |
- word2vector
| model | acc(on test) | acc(on dev) | macro-f1(test) | macro-f1(dev) |
|---|---|---|---|---|
| atae_lstm | 0.6704 | 0.6685 | 0.6351 | 0.6347 |
| tsa | 0.6710 | 0.6792 | 0.6296 | 0.6445 |
- glove
| model | acc(on test) | acc(on dev) | macro-f1(test) | macro-f1(dev) |
|---|---|---|---|---|
| atae_lstm | 0.6461 | 0.6754 | 0.5863 | 0.6226 |
| tsa | 0.6567 | 0.6798 | 0.6198 | 0.6461 |
screenshots --> outputs file
.
├── ckpt
| ├── car saved model files
| ├── laptop
| └── others
├── data
│ ├── car preprocessed data files
| ├── laptop
| └── others
├── xxx.csv csv files of pre & rl labels
├── config.py
├── data_loader.py
├── layers.py
├── models.py
├── preprocess.py
├── train.py
└── utils.py
car : char level / laptop : word level