Review Sentiment Analyzer: Bulk Sentiment & Topic Tags

Review Sentiment Analyzer

Add a sentiment score, a sentiment label, topics and defect flags to reviews you already have. Paste them in, or point the Actor at any Apify dataset (for example the output of a review scraper you ran), and get one enriched row per review plus an aggregate summary.

This Actor does not scrape any website. It only processes the data you give it.

What it does

• Sentiment score and label for every review: a score from -1.0 to +1.0 and POSITIVE, NEUTRAL or NEGATIVE.
• Topic tags: Quality & Build, Price & Value, Shipping & Delivery, Durability & Defects, Customer Service, Usability.
• Defect flag when the reviewer reports a broken or non-working product.
• Rating/text mismatch flag when stars and wording strongly disagree (5 stars with a complaint), which is useful for spotting mis-rated or suspicious reviews.
• Summary report with the sentiment breakdown, average score, average rating and sentiment per topic.
• Works with any field names: common names are auto-detected, or you can map them yourself.
• Five languages: English, Italian, Spanish, German and French.

How the analysis works

The engine is rule-based, not a large language model. It counts positive and negative terms from built-in word lists, handles simple negations ("not good", "doesn't work"), and blends the result with the star rating when one is present (60% rating, 40% text). The sentimentBasis field tells you which signals were used for each review.

What that means in practice:

• Results are deterministic, fast and cheap: thousands of reviews take seconds.
• Sarcasm, irony and subtle wording are not understood. Reviews with no rating and no known sentiment words come back as NEUTRAL with sentimentBasis: "none".
• Languages other than the five listed are scored from the rating only.

If you need nuance on every single review, use an LLM-based tool. If you need a fast, consistent first pass over a large volume, this fits.

Input

Field	Type	Description
reviews	Array	Reviews as plain strings or objects. Optional if datasetId is set.
datasetId	String	ID or name of an Apify dataset in your account. Read only.
textField	String	Field with the review text. Dot paths (review.body) work. Auto-detected if empty.
titleField	String	Field with the review title. Auto-detected if empty.
ratingField	String	Field with the rating. Numbers and strings like 4.0 out of 5 stars work. Auto-detected if empty.
idField	String	Field with the review ID, copied to the output. Auto-detected if empty.
ratingScale	Integer	Highest possible rating: 5 for stars, 10 for a 1-10 scale. Default 5.
maxItems	Integer	Stop after this many reviews. 0 means no limit. Default 1000.
keepFields	Array	Extra fields to copy unchanged to each result, for example productId.

Auto-detected field names: text (text, reviewText, review, body, reviewBody, content, comment, message, snippet, description), title (title, reviewTitle, headline, subject), rating (rating, stars, ratingScore, score, starRating, reviewRating, ratingValue), ID (id, reviewId, review_id, uuid).

Example: inline reviews

{
  "reviews": [
    { "id": "r1", "title": "Great value", "text": "Excellent build quality, arrived early.", "rating": 5 },
    { "id": "r2", "text": "Broken after three weeks. Refund refused.", "rating": 1 },
    "Not good, the packaging was damaged."
  ]
}

Example: a dataset from another Actor

{
  "datasetId": "YOUR_DATASET_ID",
  "textField": "reviewText",
  "ratingField": "stars",
  "keepFields": ["productId", "date"],
  "maxItems": 5000
}

Output

One dataset row per analyzed review:

{
  "reviewId": "r2",
  "reviewTitle": "Stopped working",
  "reviewText": "Broken after three weeks. Customer service refused a refund. Waste of money.",
  "rating": 1,
  "sentimentScore": -1,
  "sentimentLabel": "NEGATIVE",
  "sentimentBasis": "rating+text",
  "positiveTerms": 0,
  "negativeTerms": 3,
  "detectedTopics": ["Price & Value", "Durability & Defects", "Customer Service"],
  "isDefectReported": true,
  "ratingTextMismatch": false,
  "source": "input"
}

The aggregate report is saved in the run's key-value store under the key SUMMARY, and is also printed in the log:

{
  "reviewsAnalyzed": 5,
  "averageSentimentScore": 0.06,
  "averageRating": 3.2,
  "sentimentBreakdown": {
    "positive": { "count": 2, "percent": 40 },
    "neutral": { "count": 1, "percent": 20 },
    "negative": { "count": 2, "percent": 40 }
  },
  "topics": [
    { "topic": "Shipping & Delivery", "mentions": 3, "positive": 2, "neutral": 0, "negative": 1 }
  ],
  "defectsReported": { "count": 1, "percent": 20 }
}

Privacy

Only the review ID, title, text and rating are copied to the output. Reviewer names, profile links and every other field in your source data are left out unless you list them in keepFields.

Pricing

You pay per analyzed review; the current rate is shown on the Actor's pricing tab. Rows skipped because they have neither text nor rating are not charged. Use maxItems to cap the cost of a run.

Common uses

• Triage the output of a review scraper: sort by sentimentScore, filter isDefectReported.
• Track sentiment per topic over time by running the same dataset mapping on a schedule.
• Pre-filter a large review set before sending only the negative or mismatched ones to an LLM.
• Feed alerts in Make, Zapier or n8n when a run finds new negative reviews.

FAQ

Does it collect reviews from Amazon, Google or other sites? No. It analyzes data you provide. Collect reviews with a source you are allowed to use, then run this Actor on the result.

Can I read a dataset from another account? Only datasets your account can access. The Actor gets read-only access to the one you select.

My fields have unusual names. What do I do? Set textField, ratingField, titleField and idField. Nested fields use dot paths, for example review.rating.value.

Why is a review NEUTRAL when it reads as negative? The wording was probably outside the built-in word lists and there was no rating. Open an issue with the example and the lists can be extended.