Stance Detection of Turkish Twitter Accounts

Abstract

Twitter offers users an environment where they can express their opinions on any subject. As a result, user activities have become an important resource for various sociological studies. In this study, we proposed machine learning methods that utilize text-based and interaction-based features to predict the opinions of users who post messages in Turkish. We also performed zero-shot classification using ChatGPT. We labeled 4,502 accounts that posted during the Turkey Presidential election in 2018 according to the candidates they supported. In our experiments with the dataset we created, we achieved a classification accuracy of 0.831 for users.