🧠 Review Sentiment Analyzer — Themes, Trends & Complaints

Extractors stop at raw reviews. Product, CX, and reputation teams pay for the analysis — themes, complaint clusters, sentiment percentages, and trends they can act on this week.

What you get

• Themes and bigram topics ranked by mention count
• Complaint clusters that group the recurring pain points
• Sentiment split (positive / neutral / negative %) plus counts and an average sentiment score
• Time-bucketed trend data and a stable analysis ID for like-for-like delta comparisons later
• Developer-response coverage %, representative example reviews, and run metadata

Use cases by team

• Product: See which features drive praise vs. churn without reading thousands of reviews by hand.
• Customer experience: Surface the top complaint clusters for this release and track whether they shrink next release.
• Reputation / brand: Quantify sentiment movement over time with reproducible numbers you can put in a report.
• Analysts: Get auditable, deterministic scores that don't drift run-to-run — safe for regulated or repeatable reporting.

Example inputs

Analyze inline reviews

{
  "reviews": [
    {
      "review_text": "Great and fast",
      "rating": 5
    },
    {
      "review_text": "Crashes after login",
      "rating": 1
    }
  ]
}

Analyze an Apify dataset you already collected

{
  "dataset_id": "YOUR_DATASET_ID",
  "max_items": 5000
}

Tune theme extraction

{
  "reviews": [
    {
      "review_text": "Battery drains overnight",
      "rating": 2
    }
  ],
  "max_themes": 20
}

📊 Sample Output

{
  "review_count": 2,
  "sentiment_positive_pct": 50.0,
  "sentiment_negative_pct": 50.0,
  "sentiment_counts": {
    "positive": 1,
    "neutral": 0,
    "negative": 1
  },
  "average_rating": 3.0,
  "themes": [
    {
      "theme": "crashes login",
      "mentions": 1
    }
  ],
  "complaint_clusters": [
    {
      "cluster": "stability",
      "mentions": 1
    }
  ],
  "method": "ngd-lexicon-v1",
  "partial": false
}

How it works

Feed it inline reviews or a user-authorized Apify dataset ID. It scores every review with a transparent deterministic lexicon-plus-rating-prior model (ngd-lexicon-v1) — not an LLM. The same reviews in produce the same analysis out, every time: no model drift, no API key, no per-token fees. Every score is reproducible and auditable. No external review source is fetched — you bring the data, it does the analysis.

Pricing

Event	Price (FREE tier)
Actor start (apify-actor-start)	$0.05 flat
Reviews analyzed (reviews-analyzed-100)	$0.50 per 100 reviews, block-rounded

Pay-per-event with a four-tier ladder (FREE → Gold): $0.50 / $0.45 / $0.40 / $0.335 per 100 reviews. Apify applies its automatic plan discount on top. Worked example: 250 reviews on FREE = $0.05 start + 3 × $0.50 = $1.55. Blocked or zero-result runs do not intentionally charge. New to Apify? Start free.

Run it — API, CLI, MCP

API:

curl -X POST "https://api.apify.com/v2/acts/nexgenwatch~review-sentiment-analyzer/runs?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" -d @input.json

CLI:

$apify call nexgenwatch/review-sentiment-analyzer -i @input.json

MCP: expose nexgenwatch/review-sentiment-analyzer through Apify's MCP server, select it as a tool, and pass the same input object.

Related actors — App Intelligence cluster

Actor	What it does
Apple App Store Reviews	Scrape public Apple App Store reviews for any app.
Google Play Reviews	Scrape public Google Play reviews for any app.
App Rankings Tracker	Track Apple App Store top-chart rankings by country.
ASO Keyword Tracker	Track keyword rankings across App Store & Google Play.

Compliance & honest limits

• You supply the review data (inline or an Apify dataset you authorize); this actor does not scrape a review source itself.
• Deterministic lexicon model — powerful for reproducibility, but it is not a generative summarizer and will not write freeform narratives.
• A source-confirmed empty input returns SUCCEEDED-0: GENUINE_EMPTY; truncated runs are labeled PARTIAL.

FAQ

Is this AI or an LLM?

No — deliberately. It is a deterministic lexicon-and-rating-prior model (ngd-lexicon-v1): same reviews in, same analysis out, with no model drift and no API key. That reproducibility is the point — it suits regulated, comparable, or repeatable reporting where an LLM's run-to-run variation is a liability.

Where do the reviews come from?

You provide them — paste them inline or pass a dataset_id from a review scraper (for example our Apple App Store or Google Play reviews actors). Nothing external is fetched.

How does pricing work?

Pay-per-event: a $0.05 start plus $0.50 per 100 reviews analyzed on FREE (lower at Bronze / Silver / Gold). All-in — no separate LLM or API-key cost.

Does zero mean blocked?

No. A confirmed-empty input is SUCCEEDED-0: GENUINE_EMPTY; only exhausted retries fail, and truncated runs are labeled PARTIAL.

Can I compare two batches over time?

Yes. Each run returns a stable analysis ID and deterministic period buckets so a later delta layer can compare like-for-like batches.

What languages does it handle?

The lexicon is tuned for English review text; other languages score but with reduced theme quality.

Can I control how many themes come back?

Yes — set max_themes (default 12, up to 50).

Troubleshooting

• Empty output? Confirm your reviews array is non-empty or that dataset_id points to a dataset you own; a genuinely empty source returns GENUINE_EMPTY by design.
• Dataset not read? The dataset ID must belong to your account and contain review-shaped objects; check max_items isn't set below your row count.
• Themes look thin? Raise max_themes, and make sure review_text (not just rating) is populated.
• Numbers differ from last run? They shouldn't — identical input yields identical output. If they differ, your input changed.
• Need sentiment-shift alerts? That delta layer is Phase 2; today's buckets + analysis ID are the foundation for it.

About

Part of the NexGenData App Intelligence cluster — a closed web of review, metadata, ranking, and keyword actors that share one honest, deterministic contract. Built by NexGenData. Questions or need a custom feed? Open an issue on the Actor page.