How Sentiment Is Worked Out

What sentiment is

Every review or comment with written text is labelled positive, neutral or negative. The label is read from the words, not the stars, which is the point: a 4-star review that spends three sentences on a complaint is not a positive review, and a 2-star rating with no text tells you nothing you did not already know from the rating.

Stars and sentiment disagree sometimes

That is normal and usually informative. Common cases:

• A high rating with a negative note, often "great product, terrible delivery".
• A low rating written politely, which reads neutral.
• Sarcasm, which is the hardest case and the one most likely to be labelled wrongly.

Where they disagree, trust the stars for your average and the sentiment for what to read first.

Items with no sentiment

Two reasons an item has no label:

• No text. A star-only rating has nothing to analyse.
• It predates analysis. New reviews are analysed as they arrive, but a large first import can bring in more history than the initial pass covers.

Analysing your history

When unanalysed reviews with text exist, a banner appears above your reviews showing how many. On Growth and above, the button next to it runs the analysis over the backlog and reports progress as it goes. It is safe to leave the page: the work continues.

On Free and Starter the banner shows the count and points at the upgrade instead, because historical analysis is a paid feature.

Where sentiment is used

• Filtering the Reviews list down to negative items, which is the fastest daily triage there is.
• The sentiment charts in Analytics.
• The review.negative webhook, if you push events into Slack or a ticketing system.

Sentiment never triggers a reply on its own. Auto-reply decides from the star rating, so a mislabelled sentiment cannot cause a wrong reply to be sent.