This repository contains code for the research on the usage of discursive repertoirs in the classification of online texts. The aim of the research was to identify if the information on discursive repertoirs can help in improving the perfomance on BERT based models in other tasks such as subjectivity and irony detection.
Each folder contain code for different tasks:
- Repertoirs per text. The code for creating a model for multi-label prediction of discursive repertoirs in texts.
- Sentipolc one task. The model used for subjectivity and irony detection on single tasks. The dataset is taken from the Sentipolc challenge.
- Sentipolc all tasks. The model used for multi-label prediction of all tasks from the Sentipolc challenge.