Expanding Latent Ideology Estimation via Machine Learning for User Classification in Online Social Media

Predicting political ideology from social media is an important task in computational social science. Many users do not openly state their political views, so their ideology must be inferred from online behavior. This work presents a machine learning approach that combines tweet text with information from retweet networks to predict latent user ideology. Textual features capture what users say, while network features capture who they interact with and how influence spreads. Supervised models are used to predict ideology scores and ideological classes. The proposed approach allows ideology information to propagate through social connections and improves prediction for users with limited labeled data. This study highlights the value of combining text and network features for understanding political behavior online.