AI glossary Conversational
What is sentiment analysis?
← All glossary termsThe use of AI to identify the tone or emotion in text, such as positive, negative or neutral, and often what it refers to.
Explained
Sentiment analysis classifies the attitude expressed in a piece of text: whether a review, a message or a survey answer is positive, negative or neutral, and sometimes which emotion it shows, such as frustration or satisfaction. More detailed forms, known as aspect-based sentiment analysis, link each opinion to what it refers to, for example a customer who likes the product but complains about delivery. It started with word lists and classic machine learning; today language models handle irony, context and mixed opinions better. Businesses use it to track customer feedback, social media, brand perception and service quality at scale.
How it works
- Lexicon-based: counts positive and negative words from a dictionary; simple, but misses context and irony.
- Machine learning: models trained on labelled examples learn which patterns signal each sentiment.
- Language models: an LLM is asked to classify the text, often with a short definition of each label and a few examples, and can also explain why.
- Aspect-based: the text is split into topics (price, staff, delivery) and each gets its own sentiment.
What it is used for
- Customer feedback: reviews, surveys, open answers in NPS forms.
- Contact centre: spotting frustrated customers in chats, emails or call transcripts so they can be prioritised.
- Social listening and brand monitoring.
- Product teams: finding which features generate complaints or praise.
- Employee feedback, with care for privacy.
Limits to keep in mind
Sarcasm, irony, mixed opinions and domain-specific language still cause errors. A phrase such as 'unpredictable' can be positive for a film and negative for a car.
Aggregate trends are usually more reliable than single labels, so it works best for spotting patterns and prioritising cases rather than for judging an individual message on its own. Checking a sample by hand regularly shows whether the labels still match how people actually write.
Frequently asked questions
What is an example of sentiment analysis?
Reading thousands of hotel reviews and showing that comments on cleanliness are mostly positive while comments on check-in times are mostly negative, so the hotel knows where to act first.
Can ChatGPT do sentiment analysis?
Yes. Large language models can classify sentiment and emotion with instructions alone, handle longer and more nuanced text, and explain their label. For large volumes, a smaller dedicated model can be cheaper and faster.
What is aspect-based sentiment analysis?
It identifies the sentiment for each aspect mentioned in a text rather than one score for the whole. 'Great food, slow service' becomes positive for food and negative for service.
How accurate is sentiment analysis?
It depends on the domain, the language and the method. It works well on clear opinions and struggles with sarcasm and mixed views. Measuring it on a sample of your own labelled data is the only reliable way to know.
Decision guides that use this term
How to put it to work
Where it shows up at Thinkia
Related terms
Sources
- Pang, Lee and Vaithyanathan · Thumbs up? Sentiment classification using machine learning techniques (arXiv)
- Jurafsky and Martin · Speech and Language Processing (3rd ed. draft)
- Pontiki et al. · SemEval-2014 Task 4: Aspect based sentiment analysis (ACL Anthology)
Updated: 2026-10-21