Sentiment analysis is the NLP task of classifying the attitude expressed in a piece of text as positive, negative or neutral. The point is not the subject but the stance taken toward it. Some implementations reduce that to three classes, some spread it across a scale like a star rating, and some split it into discrete emotions such as anger, satisfaction and disappointment.
Early approaches used word lists. "Great" counted as positive, "terrible" as negative, and the scores were added up. That method breaks quickly on negation and irony. Current models encode the sentence with its context, so they can separate the two halves of "delivery was fast but the item arrived broken". Aspect-based sentiment analysis goes further and attaches the sentiment to individual features.
It earns its keep wherever feedback piles up. Product reviews, support tickets, open-ended survey answers, social mentions and app store ratings are the common sources.
A concrete case: a brand's app store rating looks flat across three months. Split the reviews by aspect and sentiment analysis shows negative comments about the checkout step have doubled while delivery comments improved. The average was hiding two opposite movements.
Sarcasm, idiom and regional usage still throw models off, so output gets sampled and checked by hand on a regular schedule.
