Targeted Aspect Based Sentiment Analysis
The goal of this project is to predict the aspect and the sentiment of a given text and a defined target in the text
Hypothesis
By converting the given text into sentence pair QA dataset will help in Natural Language Inference (NLI) and summarizing the classifcation results can help in understanding the aspect
Utilizing BERT for Aspect-Based Sentiment Analysis via Constructing Auxiliary Sentence
Dataset
Sentihood - Dataset for targeted aspect-based sentiment analysis (TABSA), which aims to identify fine-grained polarity towards a specific aspect. The dataset consists of 5,215 sentences, 3,862 of which contain a single target, and the remainder multiple targets.
Technology Stack
PyTorch Transformers
Install dependencies
pip install -r requirements.txt