Review Sentiment Classifier: keep the nuance
Classify reviews as Positive, Neutral, Negative, or Mixed. Separate balanced criticism from uncertainty and preserve the reviewer’s context.
Free preview limits apply. Sign in to use your credits and keep a decision history.
Your next decision starts here.
Add context above, or try an example to get started.
Mixed sentiment is evidence on both sides
A review can praise the product and criticize delivery in the same sentence. Treating it as only positive or negative loses a useful distinction. Mixed should describe explicit positive and negative evaluations, while Neutral covers a lack of clear evaluative direction. Decide whether you are classifying the entire experience or one aspect, such as the product itself, before running the tool. Changing that scope changes the meaning of a correct label even when the review text stays identical.
Work through the cases
Illustrative inputs and review guidance, not recorded model outputs. Run your own context to inspect a decision.
Good product, slow delivery
The headphones sound excellent, but shipping took three weeks and support never replied.
For the whole purchase experience, this contains explicit praise and criticism, which fits a Mixed policy. If you only want product sentiment, delivery and support are separate aspects. Keep that scope explicit instead of expecting one overall label to answer all three questions.
A restrained evaluation
It is fine, pretty much what the description said.
This can be a modest endorsement or a neutral description depending on context and the annotation policy. Do not infer strong disappointment from restrained wording alone. Ask reviewers to document the same boundary so a disagreement is not confused with a model failure.
A review that says almost nothing
???
There is no explicit positive or negative evaluation here. Neutral may be the annotation convention, but it must not be interpreted as customer satisfaction. Keep insufficient-information cases identifiable in reporting, especially if negative labels trigger customer outreach.
Set the boundaries before you classify
Suggested review rules for these options. Include the rules relevant to your team in the question or context when you run a decision.
| Choice | Use it when | Check the boundary |
|---|---|---|
| Positive | An explicit favorable evaluation within the chosen scope. | Check for meaningful criticism before reducing the whole review to praise. |
| Neutral | Descriptive content without a clear favorable or unfavorable direction. | Insufficient information is not proof that the customer feels neutral. |
| Negative | An explicit unfavorable evaluation within the chosen scope. | An isolated negative word may be negated or quoted. |
| Mixed | Both explicit positive and explicit negative evaluations. | Ambiguity alone does not establish sentiment on both sides. |
Where to take a closer look
Sarcasm and missing context
“Wonderful, it broke again” needs more than a positive-word count. Short slang, emoji, and ironic reviews should be represented in your reviewed sample. Preserve the original text when documenting disagreements.
Stars answering the question for the text
If the goal is text-only sentiment, keep the star rating out of the input and compare it with the result afterward. Include stars only when your declared task is to assess the combined review signal; do not call that a text-only evaluation.
Use different review policies for reporting and outreach
For aggregate reports, inspect per-category errors and the proportion of Mixed reviews. Systematic mislabeling can bias a trend even when individual errors seem minor. Compare like-for-like product areas and language groups before interpreting a change.
For customer outreach, review the complaint itself and whether contact is appropriate instead of triggering a message solely from a Negative score. Select confidence thresholds using labeled reviews from your audience; there is no universal cutoff that makes every outreach decision reliable.
A Jev API example for your next integration
Call the iajev REST API with cURL. Receive a choice, probabilities, and confidence as JSON.
curl https://iajev.com/api/v1/decisions/run \
-H 'Authorization: Bearer YOUR_API_KEY' \
-H 'Idempotency-Key: YOUR_UNIQUE_REQUEST_ID' \
-H 'Content-Type: application/json' \
--data '{
"state": {
"message": "The app is easy to use and the reports save me hours every week, but it crashes when I export larger files and support has been slow to respond."
},
"decision": {
"type": "choice",
"question": "What sentiment does this customer review express?",
"choices": [
"Positive",
"Neutral",
"Negative",
"Mixed"
]
}
}'Questions about this tool
How do I distinguish Mixed from Neutral?
Look for evidence of both approval and criticism in the chosen scope. A sentence that says little is not Mixed just because it is hard to classify. Document how you handle extremely short reviews so Neutral does not hide a large unknown group.
Should multilingual reviews be translated first?
Compare reviewed samples in their original language with your translation workflow before choosing. Translation can change irony, intensity, or local phrasing. Keep the original review and evaluate each language separately rather than reusing English results as evidence for another language.
Using iajev
Add context
Include the facts needed for this decision.
Review the choices
Customize the question and options to match your workflow.
Run and review
Inspect the distribution before taking action.
Is confidence the probability that a decision is correct?
No. Confidence is a separate model estimate; the choice probabilities describe the distribution over your options. Neither is a measured accuracy guarantee.
Can I compare two inputs?
Edit the context and run the same question again to compare probabilities. Changing the question or choices starts a new comparison.