This project contains PyTorch implementaions for the report Exploring Aspect-Based Sentiment Analysis through Bi-LSTM Networks with Novel Attention Mechanisms.
The project involves a comparative study of three BiLSTM architectures with novel attention mechanisms for Multi-Aspect Multi-Sentiment classification based tasks. MAMS is a challenge dataset for aspect-based sentiment analysis (ABSA), in which each sentences contain at least two aspects with different sentiment polarities. The three networks explored are:
- Concatenating Sentence and Aspect in Input Layer
- Separate Sentence and Aspect with Cross Attention
- Separate Bi-LSTM and Bidirectional Cross Attention on Sentence and Aspect
"A Challenge Dataset and Effective Models for Aspect-Based Sentiment Analysis", Qingnan Jiang, Lei Chen, Ruifeng Xu, Xiang Ao, Min Yang. (EMNLP-IJCNLP 2019) [paper] [data]
A detailed Jupyter Notebook is provided for data exploration, preprocessing, model building, and evaluation.
- Alexander Turner
- Trieu Huynh
- Amy Wing Tung Hung
MIT