Sentiment Analyzer
Is it positive, negative or neutral? Every sentence scored, with the words behind it.
Each sentence is scored separately. Nothing leaves your device.
VADER only knows English words, emoticons and emoji. Most of this text is in another script, so its scores will mostly be neutral.
Sentences by tone
Every sentence
About the Sentiment Analyzer
Paste a review, comment, e-mail or post — or a whole list of them, one per line — and see how positive or negative it is. Every sentence (or line) gets VADER’s four scores: the positive, neutral and negative shares of its words and a compound score from −1 (most negative) to +1 (most positive). The words that carried the sentiment are listed under each sentence, so you can see why it scored as it did.
VADER (Hutto & Gilbert, 2014) is a published, rule-based method built for social-media text: a lexicon of over 7,500 words, emoticons, emoji and slang rated by people, plus rules for negation (“not good”), intensifiers (“very good”), capitals (“GOOD”), exclamation marks and “but”. This is a faithful port that gives the same scores as the authors’ own code. It runs on your device, so feedback exports and private messages stay private.
How to use it
- Choose Sentences of one text to score each sentence of a review or message, or One text per line to score a list (survey answers, comments, reviews) line by line.
- Paste the text, or use Open file (.txt, .docx, .pdf, .odt, .rtf, .html, .md). Nothing is uploaded.
- Read the overall tone — the average compound score — and the split into positive, neutral and negative sentences or texts.
- Look at each sentence’s scores and its sentiment words; use Order to put the most negative feedback first.
- Download CSV for a spreadsheet with every score, or copy or download a plain-text report.
Examples
The build quality is excellent and it boils water really fast! Sadly, the lid feels flimsy and the handle gets far too hot. Customer support answered my question within an hour, which was great.
Positive +0.61 · Negative −0.42 · Positive +0.78 With two neutral sentences, the whole sample averages +0.19: positive overall (2 positive, 2 neutral, 1 negative).
In the second sentence only “Sadly” is in the lexicon: “flimsy” is not, which is why the score is mildly negative.
The book was good. · The book was only kind of good. · At least it isn’t a horrible book. · Not bad at all
+0.4404 · +0.3832 (“kind of” dampens) · +0.4310 (negated negative) · +0.4310
These are the authors’ own examples, and this tool returns the same values as their code.
Common uses
- Sorting product reviews, app-store comments or survey answers so the most negative ones are read first.
- Checking the tone of an e-mail, announcement or social post before you send it.
- Exporting per-line sentiment scores to a spreadsheet for a quick report on customer feedback.
- Teaching how rule-based sentiment analysis works, with every score explained word by word.
How VADER scores a sentence
Each word is looked up in the VADER lexicon, whose entries were rated by 10 people on a scale from −4 (extremely negative) to +4 (extremely positive) — “okay” is +0.9, “good” +1.9, “great” +3.1 and “horrible” −2.5. The rules then adjust these values:
- Negation — a word such as not, never or a word ending in n’t up to three words before flips and dampens the valence (× −0.74).
- Degree words — boosters (very, extremely) add 0.293 and dampeners (slightly, kind of) subtract it, a little less when they are two or three words away.
- Capitals — a sentiment word in CAPITALS adds 0.733 when the rest of the sentence is not all capitals.
- “but” — words before it count half, words after it one and a half times.
- Punctuation — each exclamation mark (up to four) adds 0.292, and several question marks add emphasis too.
The adjusted values are summed and normalised: compound = sum ÷ √(sum² + 15). The positive, neutral and negative scores split the sentence into three shares that add up to 1, also after the rules: each neutral word counts 1 and each sentiment word 1 plus its strength (so the “good” of “not good” counts on the negative side), and the extra emphasis of ! and ? goes to the larger side.
Reading the scores
The VADER authors give these typical thresholds, used here for the labels: positive when the compound score is 0.05 or more, negative when it is −0.05 or less, neutral in between. The overall tone is the average compound score of the sentences (or lines). The compound score is the one to compare; the positive, neutral and negative shares show how much of a sentence leans each way, so a long, mostly factual sentence has a high neutral share even when its compound score is clearly positive.
Source and changes
Hutto, C.J. & Gilbert, E.E. (2014), VADER: A Parsimonious Rule-based Model for Sentiment Analysis of Social Media Text, Eighth International Conference on Weblogs and Social Media (ICWSM-14). Lexicons and rules from the authors’ vaderSentiment code (MIT licence, github.com/cjhutto/vaderSentiment, commit 44fc044). Two clean-ups run first, because the original misses sentiment in such text: curly apostrophes are read as straight ones (so “don’t like” is negative), and emoji variation selectors and skin-tone marks are ignored (so “❤️” counts like “❤”).
Limitations
- English only: the lexicon has English words, slang, emoticons and emoji. Hindi, Hinglish and other languages are mostly scored as neutral.
- VADER scores sentiment words, not situations. A complaint without such words — “The refund took three weeks and I had to call five times” — scores neutral.
- Sarcasm, irony and context (“great, another delay”) are not understood, and domain words that are positive in one field and negative in another are scored the same everywhere.
- Up to 20,000 sentences or lines are analysed at a time; the page lists the first 300, and the CSV and the report include all of them.
Privacy
Everything happens in your browser. What you enter or open here is not uploaded or stored by MySmartCoPilot. The VADER word list (about 140 KB) is downloaded from MySmartCoPilot the first time you analyse a text; opening a PDF downloads the PDF engine once. Your text never leaves your device.
Frequently asked questions
What does the compound score mean?
It is VADER’s single overall measure for a sentence: the sum of its word valences after the rules, scaled to between −1 (most negative) and +1 (most positive). Scores of 0.05 or more are usually read as positive, −0.05 or less as negative.
Why is a clearly unhappy review scored as neutral?
Because it contains no words from the sentiment lexicon. VADER finds sentiment in words such as “bad”, “slow” or “disappointed”, not in facts such as “it took three weeks”. Read neutral feedback too, especially in complaints.
Should I use “Sentences of one text” or “One text per line”?
Use Sentences for one review, e-mail or article: each sentence is scored and the overall tone is their average. Use One text per line for a list of separate texts — survey answers, comments, tweets — so each line is scored as a whole, like a single post.
Are the scores the same as the Python VADER library?
Yes, for the same text. The rules are ported line by line and checked against the original code: 125 chosen test sentences, including the authors’ examples, and 600 random ones built to exercise every rule give exactly the same four scores. The only differences are the two clean-ups described above (curly apostrophes and emoji modifiers).
Can I analyse a CSV export of reviews?
Paste the column of reviews (one per line) into the box in One text per line mode, then Download CSV for every line with its tone and scores, ready for Excel or Google Sheets. A text that starts with =, +, - or @ gets a leading apostrophe in the file, so the spreadsheet shows it as text instead of running it as a formula.
Is my text sent anywhere?
No. The analysis runs in your browser; only the VADER word list is downloaded from MySmartCoPilot once. Your text and files stay on your device.