A new paper explores the sensitivity of computational text-based ideal point estimation (CT-IPE) methods to hyperparameter choices. The research, which involved 17 algorithms and over 4.25 million position estimates, suggests that these methods are better understood as configurable pipelines rather than fixed estimators. Analyses indicate that hyperparameter profiles explain little residual variance, with the selection of language or embedding models, seed keyword lists, and topic numbers being the most consequential researcher decisions. AI
IMPACT This research provides insights into the variability and reliability of computational methods used in political science, potentially impacting how text analysis is applied in social science research.
RANK_REASON The cluster contains an academic paper detailing a comparative experiment and analysis of algorithms. [lever_c_demoted from research: ic=1 ai=0.7]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- Influence Flower
- Patrick Parschan
- ScienceCast
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →