In this dataset each row is paragraph of text. The text is a market sentiment that might have been published in the news media or the anual busniess report. Each of these texts is qualified as either "positive", "negative" or "neutral".
