What is sentiment?

A score for whether conversation about a brand is positive, negative or neutral.

Sentiment analysis sorts posts and comments into positive, negative and neutral, usually with software, then rolls them up into a score or a percentage. It’s a quick way to spot a sudden change, like a product problem or a backlash.

It struggles with the way many audiences talk. In a real Arcus audience report for a streaming service, the most-liked reply under one of the service’s own posts was “This looks bad. I’ll be watching.” with 16,481 likes. Read literally, it’s negative. In that room it means “I’m in.” Sarcasm was only 1 in 36 of the most-liked comments, and it was the second most-liked kind when it appeared. A sentiment score can turn that kind of support into a warning.

Why it matters

Sentiment is the headline of most listening reports. Automated scores miss sarcasm and in-jokes, which is how many audiences talk.

An example

Example (made-up numbers): a listening tool scores a trailer’s comments 60% negative. Read by a person, a third of the “negative” comments are fans joking that they can’t wait.

Where it misleads

Treating a sentiment score as a measure of whether people will buy, watch or sign up. Tone and intent are different things, and the score rarely says why it moved.

How Arcus handles it

We label conversation by what people are doing (joking, praising, complaining, asking), with a person checking the labels, and show real lines with their counts. Sentiment is one column in the answer, never the headline.

More in our guide: Social listening that counts, not guesses