Sentiment analysis

A natural language processing technique that classifies a piece of text as positive, negative, or neutral — useful as a routing signal, dangerous as a verdict.

Sentiment analysis is an NLP technique that assigns a polarity score — usually positive, negative, neutral, or a number on a -1 to +1 scale — to a piece of text. Applied to customer feedback, it lets a team summarize tens of thousands of messages without reading each one: 64% positive this week, down from 71% last week, driven by support tickets about exports.

Sentiment scores are useful as a directional signal and dangerous as a primary metric. They work for monitoring tone trends and triaging volume — "show me the worst-sentiment tickets first" is a reasonable filter. They fail at nuance: sarcasm, mixed sentiment in the same message, and context-dependent meaning all confuse the simpler classifiers, and even LLM-based scoring drifts in ways that make month-over-month comparisons unstable.

For product decisions, sentiment is best used as a routing signal — finding the painful clusters worth investigating — rather than as a verdict. A 7%-drop-in-sentiment headline is at most a question worth asking; the decision still rests on reading the underlying signals.