This repository contains code for aspect-based sentiment analysis (ABSA) for the "restaurant" domain. The task is to find the sentiment polarity (positive, negative, neutral) of a given sentence corresponding to the aspect term. For instance, consider the review:- "The appetizers are ok, but the service is slow". This review/sentence has 'positive' polarity for aspect 'taste '. The polarity is 'negative' for aspect 'service.'
The word2vec and glove are excluded from the repository and have to be download separately. baseline.ipynb
uses word2vec and glove word-embeddings.
- Experiment with different word-embeddings approaches.
- Experiment with architectures such as Hierarchical Attention models using Bi-GRUs, CNNs.
- Experiment on different standard datasets in aspect-based sentiment analysis such as "laptop customer reviews".