Sentiment analyser
Score text with a local multilingual review-rating model.
Sentiment analyser
How to use Sentiment analyser
Sentiment analyser scores text with bert-base-multilingual-uncased-sentiment, a multilingual review-rating model that runs in your browser. It returns the model's five rating labels as JSON, each with a score and ranked from most to least likely. Paste 1 to 8,000 characters. Scores reflect review-style ratings, not a psychological assessment, and language, sarcasm and domain affect results.
- Paste the source into the input editor, or load the built-in example.
- Check the source format before running the operation.
- Run sentiment analyser, review the output, then use the available copy or download controls.
What this tool supports
Downloads an inference model on first run. Chat and generation require WebGPU; smaller recognition models use browser WebAssembly. AI output can omit or invent details. Review it before use.
Limits and processing
Text sources are capped at 8,000 characters; image search accepts up to 20 images and a 500-character query. Models may truncate long text. First downloads can be hundreds of MB and need browser memory.
Example source
A useful tool makes everyday work easier. Review the results before sharing.
Frequently asked questions
Does the sentiment analyser work in languages other than English?
It uses a multilingual review-rating model, so it accepts text in several languages. Accuracy still depends on the language and subject, and the model scores text as a review rating, so results on other kinds of writing need interpretation.
How do I read the sentiment analysis output?
The output is a JSON list of the model's five rating labels, each with a score between 0 and 1, highest first. The top label is the most likely rating; close scores between neighboring labels suggest mixed or uncertain sentiment.
Can sentiment analysis detect sarcasm?
Not reliably. The tool notes that sarcasm, language and domain affect results, because the model predicts a rating from the wording. Treat a surprising score as a reason to reread the text, not as a definitive judgment.
Where is my input processed?
Local mode downloads public model/runtime assets and keeps input in this browser. Models may remain cached on this device.
What are the input limits?
Text sources are capped at 8,000 characters; image search accepts up to 20 images and a 500-character query. Models may truncate long text. First downloads can be hundreds of MB and need browser memory.