MSc Dissertation

March – September 2020

In collaboration with Emotech

Department of Computer Science,
UCL,
London, United Kingdom

This project involved the application of the Albert language model to the problem of aspect-based sentiment analysis (ABSA). We adapted an adversarial training framework for ABSA tasks, substituting the smallest Albert model for larger and slower BERT-based models. We then trained the model using PyTorch using eight publicly available ABSA datasets covering two languages.

Read more here.