Glossary · Signal

What is sentiment analysis?

Sentiment analysis is the automatic classification of a piece of text by the attitude it expresses — usually as positive, negative or neutral, sometimes on a numeric scale or across finer emotional categories.

Start freeRead the glossary
[01] THE LONGER ANSWER/ SIGNAL

Why sentiment analysis matters

It is the oldest AI feature in this category and the one most likely to be quoted without its error bars. Classifying a clearly angry review is easy and every tool does it well. Classifying the messy middle — a technical bug report, a backhanded compliment, an in-joke, a comparison that praises one thing while criticising another — is where accuracy falls off, and the messy middle is most of what a listening tool collects. The practical consequence is that sentiment is a good aggregate signal and a poor individual one.
[02] HOW IT WORKS/ IN PRACTICE

From lexicons to language models

Early systems counted words against a list of positive and negative terms. Modern ones use a language model that reads the sentence in context, which handles negation and comparison far better and still struggles with irony and domain jargon.

Three labels, or a number

Most tools output positive, negative or neutral. Some emit a score. A few classify emotions — anger, joy, frustration — which sounds more useful and is harder to get right, so the error compounds.

Neutral is doing a lot of work

Most public mentions are genuinely neutral, so a tool reporting 70% neutral is usually being accurate rather than lazy. Treat a big neutral share as normal, not as a failure of the classifier.

It aggregates better than it individuates

A single mention labelled negative might be wrong. Ten thousand mentions turning more negative week over week is a real signal even when a good share of the individual labels are wrong, because the errors are roughly random and the trend is not.
[03] WHAT IT IS NOT/ CONFUSED WITH

Sentiment analysis against the terms it gets mixed up with

Most of the confusion in this category is vocabulary rather than substance. These are the terms sentiment analysis is most often used interchangeably with, and the difference that actually matters in each case.

Buyer intent

Sentiment is how someone feels; intent is what they are about to do. An angry post about a competitor is negative sentiment and excellent intent. A delighted post about your product is positive sentiment and no intent at all. Confusing the two is the most expensive mistake in this list.

Relevance scoring

Relevance asks whether a mention is about you at all — the hard part when your brand name is also an ordinary word. Sentiment assumes relevance and asks how the author feels. Filtering on sentiment before relevance produces confident nonsense.

Net Promoter Score

NPS is a survey answer from a customer you asked. Sentiment is inferred from a public post by somebody who did not know you were reading. They measure different populations and should not be plotted together.

[04] WHERE IT MISLEADS/ THE CAVEAT

What a definition will not tell you

Ask any vendor quoting a sentiment accuracy figure what it was measured on. Accuracy on product reviews — long, explicit, written to express an opinion — is much higher than accuracy on forum comments, which are short, ironic and full of domain jargon. A tool advertising 90% is usually quoting the first and selling you the second. Use sentiment for trends and for filtering competitor keywords; do not use it to decide whether a particular person is unhappy.

What FishOut does about it

FishOut marks sentiment on every post as positive, neutral or negative, treats it as one signal rather than the headline, and lets you switch it off per project. What decides whether a post reaches your inbox is relevance, read against your Brief. The most useful combination in the product is a competitor keyword with a Complaint label or negative sentiment — the built-in Switchers view — which is a switching signal rather than a mood.

Questions

Read next

Buyer intentRelevance and labelsShare of voice

Other terms

Social listeningBuyer intentShare of voiceBrand monitoringMCP (Model Context Protocol)Generative engine optimisation (GEO)Webhook