Improve reviews language detection
completed
Mikhail Malyuta
updated the status to
completed
🚀 Massive Upgrades to Language Detection!
Our new language detection pipeline is live! Here’s how we're making review categorization more accurate than ever:
🤖 Google Play Accuracy Boost: We've bumped accuracy from 94% to 98.5%, significantly reducing mislabeled languages while maintaining 99% coverage.
🍏 App Store Coverage Maxed Out: By smartly leveraging "review country" data, we've increased our App Store coverage from 97% to nearly 100%, while also bumping accuracy to 99%.
🔤 Smarter Alphabet Matching: Language predictions now strictly match the review’s alphabet (e.g., no more Cyrillic reviews accidentally tagged as Korean).
🌏 Full Chinese Script Support: We now fully support Simplified Chinese, Traditional Chinese, and Cantonese, making your auto-responses much more reliable.
🤷 Handling Emojis & Ambiguous Words: For ambiguous reviews (like just a string of emojis or the word "wow"), we now smartly fall back to the device language or review country.
These updates are live under the hood to keep your review data clean and reliable!
Mikhail Malyuta
🚀 Update: Improved Language Detection for Google Play
We have upgraded our language detection capabilities for Google Play reviews to ensure higher accuracy!
How to unlock it:
To benefit from this improvement, simply ensure the Country Connect integration is active for your app. We now use this connection to sync confirmed language data directly from the Console.
📖 Learn how to set up Country Connect here (https://support.appfollow.io/hc/en-us/articles/22164060193297-Getting-Country-for-Google-Play-Reviews)
👀 Coming Soon: We are rolling out further detection model improvements for all stores later in Q1!
Photo Viewer
View photos in a modal
W
Worthwhile Elk
Improve review language detection to avoid AI auto-replies being published in a wrong language
Anatoly Sharifulin
updated the status to
in progress
Anatoly Sharifulin
Merged in a post:
Enhance Language Detection for Short Reviews and Transliteration Cases
G
Grateful Salmon
Customers have reported inaccuracies in the language detection of reviews, particularly in the following scenarios:
- Short reviews (less than 2 words): The language detection accuracy is low, and no language is assigned in these cases.
- Medium-length reviews (3–5 words): Misclassifications occur more frequently due to limited context.
- Reviews with transliterations: Mixed-language text can cause incorrect language identification.
Impact:
These issues affect the reliability of review analysis and create challenges for customers relying on accurate language detection for categorization, tagging, and responding.
Additional Context:
This request stems from specific customer feedback, highlighting cases where the word "bien" was detected as French instead of Spanish. Similar instances have been reported across other languages as well.
Anatoly Sharifulin
updated the status to
planned
Anatoly Sharifulin
updated the status to
under review
D
Due Capybara
Improve review language detection as we had three English reviews with the wrong language detected in one week. It affects the language used for AI auto-reply.
Photo Viewer
View photos in a modal