5,300 annotated news documents and 38,000 paragraphs containing sentiment labels for 3,200 unique target entities. The dataset captures author-specific sentiment towards a central subject across both full-text and individual paragraph contexts.
Use Cases
- Train entity-centric sentiment analysis models using the target entity and document-level sentiment annotations
- Analyze sentiment consistency across a text by comparing paragraph-level labels to the overall document sentiment
- Develop stance detection algorithms that identify an author's bias towards specific subjects in news reporting
Strengths
- 5,300 documents and 38,000 paragraphs annotated for sentiment
- Covers 3,200 unique target entities across news media
- Provides sentiment labels for both the main entity at the document level and individual paragraphs