Overview
This article documents a comprehensive examination of customer reviews for Xente — a payment and shopping platform in Uganda. Reviews were scraped from Google Play Store and analysed using R.
Data Source
628 reviews obtained from Google Play Store using web scraping (Beautiful Soup). The author notes the same objective could be achieved using Rvest — a tidyverse package for web scraping in R.
Key Findings
Word Frequency Analysis
Most common words (excluding stopwords and app-name terms) revealed user concerns around functionality and experience.
Sentiment Analysis
Using Bing and NRC lexicons:
- ›Negative drivers: "invalid" and "error" dominated sadness-related sentiments
- ›Positive words: "love" (avg 5-star rating), "awesome" (4.88 stars)
- ›Negative words: "slow" (1.17-star average), "fake" (1.38 stars)
- ›Trend observation: Trust sentiment showed decline during the observation period — worth tracking
Bigram Networks
Visualised word relationships revealed notable pairs including "mobile money" and patterns centred around "app" and "airtime."
Negation Analysis
Examined words preceded by "not," "no," "never," and "without" to understand reversed sentiments — a common source of error in simple sentiment models.
Word Cloud
Generated visual frequency distribution of review vocabulary.
Key Takeaways
1. Users appreciate ease of use and functionality (positive: "easy," "nice," "cool")
2. Technical issues frustrate customers the most (negative: "error," "invalid," "slow")
3. Trust metrics declined during the observation period — a signal worth investigating
4. The company actively responds to customer feedback