Review Theme Analyzer

Your review scraper hands you rows. This Actor hands you the story inside them.

Point it at any review dataset, for example the output of a Google Maps, Amazon, TripAdvisor, or app store review scraper, or paste reviews directly. It reads every review and returns what customers actually praise and what they actually complain about, ranked, counted, and quoted.

What you get

One summary record with the totals: reviews analyzed, positive and negative counts, average rating, and the single biggest praise theme and complaint theme.

One record per theme such as Food and drink, Service and staff, Speed and waiting, Cleanliness, Price and value, Quality and durability, and Delivery and packaging. Each theme record carries its review count, its share of all reviews, its praise or complaint bucket, its average score and rating, and up to three representative quotes taken word for word from your reviews. Frequent two word phrases from your own reviews also surface as themes, so product specific topics such as battery life appear even though they are on no fixed list.

One breakdown record per group when your records carry a group field such as a location, a store, or a product name. Each group gets its own counts, average rating, and top praise and complaint themes, so you can see which location struggles with waiting times and which one wins on service.

Why it is different

Sentiment tools on the Store label one comment at a time, or scrape a single profile and label what they find. None of them accepts an arbitrary review dataset and returns ranked themes. This Actor is the downstream step the big review scrapers leave undone: synthesis you can act on, in structured records you can sort and chart.

How the analysis works

The scoring is deterministic and local. A transparent word and phrase lexicon reads each review clause by clause, so a complaint about damaged packaging inside a happy review still lands on the complaint side, and each theme is judged by the words actually said about it. Star ratings anchor the overall positive and negative sorting when the dataset carries them, and negation is handled, so a review that says the room was not clean counts against Cleanliness rather than for it. The same input always produces the same output, and no review text ever leaves the run for an outside AI service.

Input

Dataset ID with reviews. The output dataset of any review scraper. Or use Inline reviews and paste a JSON array of records, or plain text strings.

Your Apify API token. Only for dataset ID input. Apify runs cannot always read a dataset on their own: on accounts with restricted storage access the platform refuses the run, and your own token is the key that opens your dataset. It is a secret input, used once to read that one dataset, and never written to the output. Inline reviews never need it.

Review text field. The field that holds the review text. Common field names are found automatically when a record leaves it empty.

Group field. Optional. Names the field that groups reviews, such as location or product.

Rating field and Max reviews to analyze. Ratings anchor the sorting when present. The cap is also a spend cap, because you are charged only for reviews actually analyzed.

Pricing

Pay per event: $0.001 Actor start + $0.002 per review analyzed. Records without usable review text are skipped and never charged, and an empty or failed input charges no review events at all. Analyzing 500 reviews costs about one dollar.

Notes

Theme finding is lexicon based, so it is fast, repeatable, and honest about its method, but it will not catch sarcasm or themes expressed in unusual wording. Quotes are shortened to 240 characters. Ratings are read from the first field that carries a value from 1 to 5 among the rating field you name and the common rating field names.