Sentiment Analysis and NLP On Google Play Store Reviews

Abstract - The Google Play Store sees thousands of new apps regularly, developed by individuals or teams competing globally. Most apps are free, making revenue models like ads and in-app purchases unclear. As a result, app success is often judged by installation counts and user ratings. However, ratings can be biased or inconsistent, and there’s often a gap between number of ratings and written reviews. This study uses machine learning to predict app ratings based on a dataset from Kaggle, analysing features such as app type, user reviews, and ratings. The analysis merges app data with user reviews and performs sentiment analysis using Sentiment Polarity. The sentiment distribution is visualized for free and paid apps using bar charts. Review texts are summarized using the Sumy library, and spam is detected by flagging reviews with fewer than three words or common spam keywords. Summaries and spam labels are saved to new CSV files for further analysis. Reviews are then cleaned, lemmatized, and stripped of stop words before being fed into a logistic regression model trained on a small labelled dataset. The model classifies reviews into topics like ’bug’, ’UI’, or ’feature request’ and is evaluated on a test set. Finally, it predicts categories for unlabelled reviews, and a pie chart shows how many summaries contain meaningful content versus empty ones. Keywords: - Google Play Store Apps, Ratings Prediction, Data Analysis Library, Sentiment Analysis, NLP, Machine Learning Algorithm