Sentiment Analyzer — Text Signals, Emotion & Keywords

Batch English-language sentiment analysis with emotion signals, key phrase extraction, and language labeling. Processes up to 100,000 texts using offline NLP without an external AI API.

The analyzer runs inside the Actor process with deterministic lexicon and rule-based scoring. It is suitable when you need repeatable signals without sending text to a separate model API.

How to Use

Zero-config (just provide texts):

{
  "texts": [
    "This product is amazing! Best purchase ever."
  ]
}

With ID and metadata:

{
  "texts": [
    { "id": "review-123", "text": "Love this product!", "metadata": { "source": "amazon" } }
  ]
}

Review analysis pipeline (filter negative reviews):

{
  "texts": ["... your review texts ..."],
  "onlySentiment": "negative",
  "extractKeyPhrases": true
}

Input Configuration

Parameter	Type	Default	Description
texts	array	sample texts	Text strings to analyze. API callers can also send {id, text, metadata} objects. Max 100K.
sentimentModel	string	"combined"	afinn, vader, or combined (most accurate)
detectEmotions	boolean	true	Detect joy, anger, sadness, fear, surprise, disgust
extractKeyPhrases	boolean	true	Extract key phrases contributing to sentiment
detectLanguage	boolean	true	Detect language of each text
neutralThreshold	number	0.05	Compound score range for neutral classification
onlySentiment	string	null	Filter: positive, negative, neutral, mixed
includeWordScores	boolean	false	Per-word sentiment scores
minimumConfidence	number	0	Minimum confidence threshold

Output Format

{
  "id": "review-123",
  "text": "This product is amazing!",
  "textLength": 25,
  "language": { "detected": "en", "name": "English", "confidence": 0.9 },
  "sentiment": {
    "label": "positive",
    "compound": 0.8721,
    "positive": 0.8721,
    "negative": 0.0,
    "neutral": 0.1279,
    "confidence": 0.9721
  },
  "emotions": {
    "joy": 0.8, "anger": 0.0, "sadness": 0.0,
    "fear": 0.0, "surprise": 0.1, "disgust": 0.0,
    "dominant": "joy"
  },
  "keyPhrases": [
    { "phrase": "amazing", "sentiment": 1.0, "importance": 0.9 }
  ],
  "analyzedAt": "2026-03-11T12:00:00.000Z"
}

Sentiment Score Interpretation

Compound Score	Label	Description
0.05 to 1.0	positive	Positive sentiment
-0.05 to 0.05	neutral	No strong sentiment
-1.0 to -0.05	negative	Negative sentiment
Both pos & neg > 0.3	mixed	Contains both positive and negative

Tips and Advanced Usage

Choosing the right model:

• combined (default) — Uses the AFINN and VADER-style signals together
• vader — Better for social media with emojis, slang, caps, punctuation
• afinn — Fastest, purely lexicon-based

Adjusting neutral threshold:

• Default 0.05 works well for most use cases
• Increase to 0.15-0.20 for stricter positive/negative classification

Handling sarcasm: Lexicon-based analysis cannot reliably detect sarcasm ("Oh great, another delay"). For sarcasm, consider LLM-based tools.

Multi-language: Works best with English. 20+ languages supported via translated lexicons with lower accuracy.

Pricing

Pay-Per-Event: $0.0039 for each text-analyzed event ($3.90 per 1,000 texts)

Pricing includes all platform compute costs — no hidden fees.

Scenario	Texts	Cost
Product reviews batch (500)	500	$1.95
Twitter mentions (5,000)	5,000	$19.50
Customer feedback (20,000)	20,000	$78.00
Monthly monitoring (100,000)	100,000	$390.00

Related Actors

• Finance News Scraper — Pair with sentiment for financial signal detection
• Google News Scraper — News to analyze with sentiment

FAQ

How accurate is the sentiment analysis?

Accuracy depends on domain, wording, irony, and language. Validate the deterministic scores against labeled examples from your own use case before automating decisions.

Does it detect sarcasm?

No. Sarcasm detection requires contextual understanding that lexicon-based NLP cannot reliably provide. Document this limitation for your users.

What languages are supported?

Sentiment scoring is designed for English text. Language detection may label other languages, but that does not make the English sentiment lexicon reliable for them.

Can I analyze social media posts with emojis?

Yes — emoji sentiment mapping is built in. Common emojis are mapped to sentiment values.

What's the difference between AFINN and VADER models?

AFINN: lexicon-based word scoring, fast and simple. VADER: rule-based with negation handling, emphasis (CAPS, !!!), and emoji support — better for social media.

How does emotion detection work?

Emotion detection uses a word-emotion association lexicon. Each word is mapped to one of 6 emotions. The dominant emotion is the one most frequently detected in the text.