Choosing-a-Classification-Method-for-Measuring-Government-Criticism-in-Armenian-Civil-Society-Texts

Summary

This repository contains a political science research project comparing machine learning methods for classifying government criticism in Armenian civil society texts. It implements and evaluates five classifiers (Naive Bayes, Logistic Regression, Random Forest, Gradient Boosting, Stacking Ensemble) using two text representations (bag-of-words and BERT embeddings) on a hand-coded dataset of 642 documents. The project then applies the best model to classify ~4,100 publications from five Armenian watchdog organizations from 2014-2024, analyzing trends in government criticism before and after the 2018 Velvet Revolution.