What is sentiment analysis?
Sentiment analysis classifies text by the attitude it expresses, usually positive, negative or neutral, and ideally by how strongly, so large volumes of reviews, tickets or comments can be read at a glance.
How does sentiment analysis work?
There are two broad approaches. Lexicon-based methods score words from a dictionary of known positive and negative terms and adjust for grammar, such as negation. Machine learning methods learn from labelled examples and predict the sentiment of new text. Both turn a sentence into a score, often between −1 and +1.
Sentiment analysis examples
- “The new export is brilliant, saved me an hour.” Strongly positive.
- “Still waiting for a reply after three days.” Negative, with urgency.
- “The update arrived on Tuesday.” Neutral.
- “Not bad, but the setup was painful.” Mixed, leaning negative.
Types of sentiment analysis
- Document level: one score for a whole review or ticket.
- Sentence level: a score for each sentence, which catches mixed messages.
- Aspect based: sentiment towards specific topics, such as billing versus the product.
- Emotion detection: labels like frustration, confusion or delight instead of just positive and negative.
What limits sentiment analysis accuracy
Sentiment analysis accuracy depends on your customers’ language, sarcasm, negation and domain words: “sick” is praise in one place and a complaint in another. For sentiment analysis for customer service, seeing why a score came out as it did matters as much as the score. Customer Support Sentiment Analyzer shows the exact terms that moved each score, so a disputed rating can be checked.
Last reviewed 2026-11-01.