I’m doing an economics research project and I’m trying to finetune a large language model to perform sentiment classification. However, my dataset is not labelled on an absolute scale, but on a relative scale compared to other data in the group. In each group, there are three pieces of texts, a neutral text, a more positive version of the text compared to the neutral, and a more negative version compared to the neutral.
What language model should I use? And how should I finetune my model?
Any help would be much appreciated.