Sentiment Analyzer

Sentiment analysis estimates the emotional tone of a piece of writing. This tool treats every line you paste as a separate item, which makes it handy for a list of product reviews, survey answers, social media comments or customer emails. Each line receives a label of positive or negative and a percentage showing how sure the model is.

The text is analysed by a model running in your browser, so feedback from customers or colleagues is not sent to any third-party service. Once the model has been downloaded, you can analyse as many lines as you like.

How to use it

1. Paste or type your English text into the box, putting each review or comment on its own line.
2. On first use, wait for the model to download; it is about 70 MB and is cached afterwards.
3. Click analyze to score every line.
4. Sort or scan the results, and read the lines with low confidence yourself, since those are often mixed or neutral.

How it works

Scores come from DistilBERT fine-tuned on SST-2 (Xenova/distilbert-base-uncased-finetuned-sst-2-english), run with Transformers.js. DistilBERT is a smaller, faster version of Google's BERT language model that reads a whole sentence at once and represents the meaning of each word in context. It was then trained on the Stanford Sentiment Treebank, a set of movie review sentences labelled by people as positive or negative. For each line, the model outputs a probability for both labels, and the tool shows the larger one.

Good for

• Getting a quick overview of a batch of product or app store reviews.
• Sorting open-ended survey responses before reading them in detail.
• Checking the tone of your own email or post before sending it.
• Demonstrating basic natural language processing in a classroom.

Limitations

• There are only two labels; neutral or mixed text is still forced into positive or negative, usually with lower confidence.
• Sarcasm, irony, slang, emojis and domain-specific language are often misread, and the model was trained on English movie reviews only.
• Very long lines are truncated, and this compact model is less nuanced than large cloud language models. Desktop Chrome or Edge gives the smoothest experience.

FAQ

More tools