FeedbackSieve

Raw app-store reviews are noisy and unstructured — a client needs signal, not a pile of unread text. FeedbackSieve fetches Play Store reviews, categorizes them, and surfaces recurring issues, grouping by topic and quantifying sentiment well enough that a product team can prioritize fixes with confidence instead of guesswork. Built and used for an education-app client's product analysis, it directly identified video playback problems from user feedback that had been buried in the noise.

What I Built

• Play Store review scraping for any target app
• Topic categorization — reviews grouped by area (video, login, performance, etc.)
• Recurring-issue detection across large volumes of feedback
• Per-category sentiment analysis to gauge severity, not just frequency

Tech Stack

• Language: Python
• Scraping: Play Store data collection
• NLP: Sentiment analysis (TextBlob / VADER / transformer-based)

Reviews get grouped by topic before sentiment analysis runs, rather than scoring the raw review stream — the client needed to prioritize by issue area, not see an aggregate sentiment number. Severity is weighted alongside frequency per category, since a small number of intensely negative reviews about video playback mattered more than a larger number of mildly mixed reviews about minor UI issues. That combination is what let the tool pinpoint video lag as the client's priority fix.