Researchers have developed a new Estonian-language dataset for document-level subjectivity analysis, comprising 1,000 texts rated on a scale from 0 to 100. Initial experiments using this dataset showed moderate inter-annotator agreement among human raters, prompting a re-annotation of divergent scores. An experiment using GPT-5 for automatic subjectivity scoring indicated feasibility but highlighted differences from human annotations, suggesting LLM-based scoring is not a direct substitute for human judgment. AI
IMPACT Provides a new resource for evaluating LLM understanding of subjective content in Estonian.
RANK_REASON The cluster contains an academic paper detailing the creation of a new dataset and an initial experiment. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →