핵심 개념
HyEnA combines human insight with AI processing to efficiently extract key arguments from opinionated texts.
초록
HyEnA, a hybrid method for extracting arguments from opinionated texts, combines human and AI capabilities. It breaks down the extraction task into annotation, consolidation, and selection phases. The method is evaluated on citizen feedback corpora related to COVID-19 measures. Annotators identify unique key arguments while maintaining coherence and diversity in the extracted arguments. The Power algorithm reduces the need for human annotation in identifying similar argument pairs. Louvain and spectral clustering methods are used to create argument clusters based on similarity labels provided by annotators. Clusters show a coherent distribution of arguments with similar stances towards policy options.
통계
HyEnA achieves higher coverage and precision than automated methods when extracting arguments from diverse opinions.
Annotators identified an average of 15 unique key arguments per option in Phase 1.
The Power algorithm reduced the number of pairs requiring human annotation by 60% in Phase 2.
Louvain clustering yielded the smallest error for two corpora, while spectral clustering performed better for one corpus.
Clusters generally represent a coherent distribution of arguments with similar stances towards policy options.