Auditor Sentiment Sentences from Financial News

Encompassing 3500 sentences from English-language financial news, categorized by sentiment. The annotations were generated by 5-8 annotators per sentence, with data divided by annotator agreement rate.

Use Cases

• Train a text classification model on 3500 financial news sentences for sentiment analysis.
• Analyze the relationship between annotator agreement rate and sentiment categories.
• Fine-tune a language model on expert-generated sentiment annotations from financial text.

Strengths

• Dataset size of 3500 annotated sentences.
• Sentiment labels generated by 5-8 expert annotators per sentence.
• Data is structured by annotator agreement rate, allowing for quality assessment.

Limitations

• Limited to 3500 sentences, which is a small sample for some machine learning tasks.
• Geographic and linguistic scope is restricted to US English financial news.
• The dataset's last update was in August 2022, potentially making it temporally stale for current financial sentiment.

Provenance

Source

News Department, collected via contact with COE for Auditing.

Collection Method

Sentences extracted from financial news and annotated for sentiment by 5-8 annotators.

Freshness

Last updated 2022-08-30.

Geography

Region is tagged as 'us', indicating US financial news.