GPT-4 Sentiment Analysis for Faroese

The dataset comprises annotations of 225 sentences extracted from 170 Faroese news articles. The analysis was conducted at both the sentence and document levels, incorporating multi-class sentiment labels. The dataset features comparisons between GPT-4’s performance and that of human annotators.

Columns

1. News article: The full text of the news article.
2. Selected Sentence: The sentence selected for sentiment analysis.
3. Sentence label - GPT-4: GPT-4’s sentiment classification of the selected sentence.
4. Sentence label - Annotator 1: The first human annotator’s sentiment classification of the selected sentence.
5. Sentence label - Annotator 2: The second human annotator’s sentiment classification of the selected sentence.
6. News label - GPT-4: GPT-4’s sentiment classification of the entire news article.
7. News label - Annotator 1: The first human annotator’s sentiment classification of the entire news article.
8. News label - Annotator 2: The second human annotator’s sentiment classification of the entire news article.
9. Topic - GPT4: GPT-4’s classification of the article’s topic.
10. Topic relevance - Annotator 1: The first human annotator’s assessment of the topic’s relevance.
11. Correct topic if not relevant - Annotator 1: The corrected topic by the first annotator if the original classification was deemed not relevant.
12. Topic (National (N) / International (I) / Mixed (M)) - Annotator 1: The topic classification as National, International, or Mixed by the first human annotator.

Release: 05.03.2024
Contact: vesteinn.snaebjarnarson@gmail.com