🔎 What was studied
This article introduces a new computational framework that detects, analyzes, and predicts partisan responsiveness — the moments when parties on opposite poles react to each other’s agendas and thereby contribute to polarization.
🗂️ What data was used
- 10,597 documents scraped from the official websites of radical right and ethnic political parties in Slovakia, covering 2004–2014.
🧭 How responsiveness was identified
- Spikes in responsiveness are detected and categorized with latent Dirichlet allocation (LDA), turning streams of text into topical events that signal inter-party reactions.
⚙️ How prediction works
- Terms that comprise the LDA topics are used as features.
- A gradient-descent solver trains a probabilistic classifier to predict whether a given issue will elicit a partisan reaction or be ignored.
- Classifier performance is evaluated against alternative methods.
📈 Key findings
- The approach predicts which political issues will trigger partisan responses with an F-measure of 83%.
- The model outperforms both Random Forest and Naive Bayes classifiers on this task.
🧾 Validation and interpretation
- Subject-matter experts validate the detection and classification approach and assist in interpreting topic-driven reaction patterns.
💡 Why it matters
- Provides a replicable, time-series text-mining method to anticipate when agenda items will escalate partisan exchange, offering a tool for scholars and analysts studying polarization dynamics.
