What is Sentiment Analysis?
Sentiment analysis is a natural language processing task that assigns an evaluative orientation to text, most often positive, negative, or neutral, and sometimes a numeric intensity or a set of emotion categories. It is applied to reviews, survey responses, support messages, and social posts. Results describe expressed language, which does not always match what a writer actually feels.
Granularity varies. Document level analysis produces one label for a whole text, sentence level produces one label per sentence, and aspect based analysis attaches a separate orientation to each topic mentioned, so a single review can be positive about delivery and negative about durability. Aspect based output is more useful operationally and considerably harder to produce reliably.
Early systems scored texts by counting words drawn from polarity lexicons, adjusted by negation and intensifier rules. Supervised classifiers trained on labeled reviews improved on this, and fine tuned neural encoders or prompted general purpose language models now handle most cases. Lexicon methods remain in use where transparency and low cost matter more than accuracy.
Accuracy is limited by phenomena that depend on context outside the sentence. Sarcasm, comparison, conditional statements, quoted complaints, and domain specific vocabulary all mislead models, and a factual sentence containing negative words is easily scored as negative opinion. Human annotators also disagree on borderline cases, which caps the accuracy any model can be shown to reach.
Because labels are aggregated into metrics that drive decisions, calibration matters more than any individual prediction. Teams monitor the distribution of scores over time, sample outputs for manual review, and avoid treating a sentiment score as a measurement of one person's state. Reporting over groups of texts is far more defensible than acting on a single classification.
Key points
- Classifies text as positive, negative, or neutral.
- Available at document, sentence, or aspect level.
- Sarcasm, context, and domain vocabulary reduce accuracy.
- Measures expressed language, not inner feeling.
- More reliable in aggregate than per message.
In practice
A team collects several thousand app reviews each month. Aspect based sentiment analysis separates comments about pricing, speed, and support, showing that speed sentiment fell after a release while pricing sentiment held steady. That pattern points engineering at a performance regression. No individual review is treated as evidence, since the useful signal is the aggregate shift.