ABSA (Aspect-Based Sentiment Analysis) is a task used to identify specific aspects within a text and assign them a polarity (e.g., negative, neutral, or positive). My project focuses on this task within a particular context: COP, the annual conference on climate change. Every year, people around the world express support or discontent regarding COP through social media platforms like X (formerly Twitter). The main goal of this project is to collect and analyse these data to fine-tune a pre-trained model for ABSA, specifically within the Climate Change and COP domain.
The project consists of two main steps:
• Step 1: Use Mistral to generate samples, which will serve as a training dataset for fine-tuning.
• Step 2: Fine-tune a pre-trained model on the ABSA task within the COP domain.
Finally, the model was evaluated using a test set, also generated by Mistral.
This project has been developed as a coursework assigned for the class "Social Network and Media Analysis", Spring 2024 (Instructor: prof. Andrea Tagarelli) at the DIMES Department, University of Calabria, Italy.