NN4ABSA
Neural Network based model for Aspect-Based Sentiment Analysis.
- NOTE: it is NOT related to our finished or ongoing research projects.
Model 1
- Word embeddings: stanford GloVe
- Ctx Feat Extractor: CNN + Multi-Channel
- Target Feat Extractor: Weighted sum of word vectors making up the target phrase
Performance (accuracy & macro-F1)
|
14semval-restaurant |
14semeval-laptop |
Twitter |
| ATAE-LSTM [1] |
77.2/- |
68.7/ |
- |
| MemNet [2] |
78.16/65.83 |
70.33/64.09 |
68.50/66.91 |
| IAN [3] |
78.6/- |
72.1/- |
- |
| RAM [4] |
80.23/70.80 |
74.49/71.35 |
69.36/67.30 |
| Model 1 |
79.43/69.49 |
74.65/69.27 |
71.10/69.32 |
References
- Attention-based LSTM for Aspect-level Sentiment Classification. EMNLP 2016
- Aspect Level Sentiment Classification with Deep Memory Network. EMNLP 2016
- Interactive Attention Networks for Aspect-Level Sentiment Classification. IJCAI 2017
- Recurrent Attention Network on Memory for Aspect Sentiment Analysis. EMNLP 2017