Fixes found by analyzing 1,000 reviews from a live run rather than from fixtures.
- Billing correctness. Google Play returns the same reviews for every storefront when the language matches. A four-storefront Android run produced 400 rows that were 100 unique reviews repeated four times — and billed all 400. Reviews are now deduplicated by ID across the whole app before charging, and Android storefronts that map to an already-covered language are skipped entirely. A measured 38% overbilling on that run drops to zero.
- Google Play storefronts now map to languages (
br → pt, de → de, …), which is the parameter that actually changes Google's review set.
- A 1- or 2-star review can no longer be labelled positive. "Respect the Goat 🇦🇷" is a protest review; AFINN scores "respect" at +2 and it came back positive. Low stars now cap the label at
mixed.
- Language detection fixed for short text. Short foreign reviews like "Ist Nice. Duo ist geil" were being scored with the English lexicon. 53 of 1,000 live reviews were rerouted to the correct language.
- Four new topics for the biggest uncovered clusters:
gamification_rewards, brand_controversy, content_moderation, effectiveness_results. Untagged reviews dropped from 55.2% to 45.6% on the same live sample.
First release.
- Reviews from the iOS App Store (public RSS feed) and Google Play, across up to 8 storefronts per run.
- Sentiment scoring in 6 languages (en, es, pt, fr, de, it) with per-review language detection.
- Negation handling — "never crashes" and "no bugs" score as praise, and a clause breaker stops the negation leaking ("never crashes but the ads are awful" stays a complaint).
- 14-topic taxonomy for app reviews, with multilingual keywords.
- Complaints ranked by pain (frequency × negativity), each with verbatim quotes.
- Per-version and per-month breakdowns, plus automatic detection of a release that regressed against the app's own baseline.
- Head-to-head comparison across 2+ apps, topic by topic, with explicit weaknesses and strengths versus the best competitor.
- Pay-per-event pricing: cheap raw reviews, separate charge for the intelligence reports.