Sentiment analysis

Classifying text by emotional tone — positive, negative, neutral, or finer-grained emotion labels — used heavily in customer reviews, social media monitoring, and market research.

Sentiment analysis is the task of categorizing text by emotional tone. The simplest version is three-class (positive, negative, neutral); finer versions detect specific emotions (joy, anger, fear, sadness, surprise) or aspect-level sentiment ("battery life is good but the screen is bad" — positive on battery, negative on screen). It matters because companies use sentiment analysis at massive scale: monitoring brand reputation across social media, scoring customer reviews, prioritizing support tickets, gauging political opinion, and tracking financial market sentiment from news. The dashboards in tools like Brandwatch, Hootsuite Insights, and Sprinklr are largely sentiment classifiers under the hood. A concrete example: e-commerce site has 100,000 reviews. Run sentiment analysis to flag the recent spike of negative reviews about "shipping delay", route them to operations, and surface representative quotes for the executive dashboard. Same data, sliced by aspect, identifies which features customers love versus complain about. LLMs handle sentiment analysis trivially with zero or few-shot prompting and outperform older specialized classifiers especially on ambiguity, sarcasm, and cross-language input. Older specialized models (BERT-based fine-tunes) are still cheaper at very high volume but increasingly being replaced by LLM API calls or distilled small models. Related: text classification, NER, prompting, evaluation.