Sentiment analysis returns numeric scores that quantify the emotional tone of text. Understanding these scores is essential for building effective features that respond appropriately to user sentiment.
The Sentiment Scale
Sentiment scores typically range from -5 (extremely negative) to +5 (extremely positive), with 0 representing neutral text. The score reflects the overall emotional valence—positive values indicate favorable sentiment, negative values indicate unfavorable sentiment. A score of +3 might represent "I really enjoy this product" while -3 might represent "This is frustrating and disappointing."
Comparative Scores
The comparative score normalizes sentiment by text length, dividing the raw score by the number of words. This allows fair comparison between a tweet and a paragraph. A comparative score of 0.5 means the text averages half a sentiment point per word—strongly positive. Scores between -0.1 and 0.1 are typically neutral.
Classification Categories
Beyond numeric scores, text is classified as positive, negative, or neutral. The thresholds vary by implementation, but typically: scores above 0.05 comparative are positive, below -0.05 are negative, and between are neutral. Some applications use five categories (very negative, negative, neutral, positive, very positive) for finer granularity.
Confidence and Edge Cases
Sentiment analysis works best on clear expressions of opinion. Sarcasm, irony, and context-dependent language may be misclassified. Mixed sentiment ("The food was great but the service was terrible") produces scores near zero despite strong opinions. Consider these limitations when designing your application logic.