Sentiment Analyzer

Sentiment analyzer for reviews, comments and messages

This sentiment analyzer reads a piece of text and tells you whether it sounds positive, neutral or negative, with a score from −1 to +1 and a rough confidence figure. It works like the well-known AFINN and VADER lexicon methods: every opinion word has a fixed valence from −5 (“horrible”) to +5 (“superb”), and the tool adds those values up while watching for negation (“not good”), intensifiers (“very”, “slightly”), capital letters, emoji, emoticons, exclamation marks and the “but” that often flips a review halfway through.

Paste one review for a sentence-by-sentence breakdown with each opinion word highlighted, or switch to One review per line to score a list of comments, survey answers or tweets at once and download the result as CSV. Everything runs in your browser; nothing is uploaded or stored.

How to use the Sentiment Analyzer

1. Paste your text into “Text to analyse”. The example text shows how mixed reviews are handled.
2. Read the overall score: above +0.05 is positive, below −0.05 is negative, and the band between counts as neutral. The dot on the bar shows where the text sits.
3. Scroll through the sentence breakdown: green words raise the score, red words lower it and yellow words were negated. Hover a word to see how it was adjusted.
4. Check the “Opinion words found” table to see which words drove the result.
5. For a batch of reviews, choose “One review per line”, paste them, and press “Download CSV” to get the score, label and confidence for every line.

Sentiment Analyzer features

How the sentiment score is calculated

The text is split into sentences, and each sentence into words, emoji and emoticons. Every token found in the lexicon contributes its valence, adjusted by the words around it:

• Negation: a negator (not, never, no, can’t, didn’t, without…) within the three words before an opinion word flips its sign and scales it by 0.75, because “not great” is weaker than “terrible”.
• Intensifiers and dampeners: “very”, “really”, “extremely” and friends multiply the score by 1.25–1.5; “slightly”, “somewhat”, “a bit” multiply it by 0.6–0.75. A booster two words away has half the effect.
• Capitals: a word written entirely in capitals, in a sentence that is not all capitals, counts 1.3 times.
• “But”: opinion words before “but” are halved and words after it are multiplied by 1.5, so “good price but terrible battery” comes out negative.
• Exclamation marks: each one (up to four) pushes the sentence 0.3 further in the direction it already leans.

From a raw sum to a −1 … +1 score

The adjusted values are added up and squashed with the same normalisation VADER uses: score = sum ÷ √(sum² + 15). A single mildly positive word gives roughly +0.4, three strong ones approach +0.9, and the score can never exceed ±1. The overall score for a text uses the sum across all sentences, so one strongly negative sentence can outweigh several lukewarm positive ones — exactly what a reader would feel.

Worked example: “The food was not good, but the staff were very friendly!” — “good” (+2) is negated to −1.5 and halved before “but” to −0.75; “friendly” (+2) is intensified by “very” to +2.6 and multiplied by 1.5 after “but” to +3.9; the exclamation mark adds +0.3. Sum = 3.45, score = 3.45 ÷ √(3.45² + 15) ≈ +0.67, labelled positive.

Where lexicon-based sentiment analysis fails

Because the tool only looks words up, it cannot understand context. Expect it to be wrong in these situations, and read the highlighted words to see why:

• Sarcasm and irony: “Great, another update that breaks everything” scores positive because “great” is in the list.
• Domain words: in medical or financial text, “positive test result” or “aggressive growth” carry different meanings; in gaming, “sick” and “killer” are compliments.
• Comparisons and conditionals: “It would be perfect if it worked” is scored as if the product were perfect.
• Negation at a distance: only the three preceding words are checked, so “I would not say that the hotel was clean” still reads as positive.
• Other languages: the lexicon is English only. Emoji still count, which is why a Spanish review with 😡 may show as negative.

The Word Frequency Counter shows which words dominate a batch of reviews, and the Text Cleaner strips quotes, signatures and HTML before scoring. For the reading level of the same text, use the Readability Checker.