A study on text mining and sentiment analysis: User comments of earthquake apps

Turkey, due to its geological structure, is a country at high risk of earthquakes. Major disasters such as the 2023 Kahramanmaraş Earthquake have clearly demonstrated the need to increase public awareness of earthquakes. Earthquakes are natural disasters with the potential to cause both loss of life and property, and people want to know about earthquakes in advance for reasons such as saving lives, reducing material damage, and being psychologically prepared. Earthquake apps on phones have become important tools for both providing early warnings and delivering earthquake-related information to users quickly and reliably. User feedback on earthquake apps they use provides crucial data for analyzing and interpreting the sentiment conveyed in the feedback. The information obtained by analyzing this data is crucial for app developers in developing apps and improving the user experience. In this study, sentiment analysis of user comments on the most commented earthquake apps was conducted using a supervised machine learning approach. Supervised machine learning classification algorithms such as Naive Bayes, Sequential Minimal Optimization, and k-nearest neighbor algorithms were used for sentiment analysis. Data collected using the Instant Data Scraper data extraction tool available on Google Play and the App Store were classified as positive, negative, and neutral. In this study, the effects of "data distribution" and "feature selection" on classification were investigated using WEKA 3.8.6 software. The study observed that an imbalanced dataset performed better than a balanced dataset. It was observed that performing feature selection while analyzing a balanced dataset did not cause a significant performance change compared to models without feature selection, and performing feature selection while analyzing an imbalanced dataset performed better than models without feature selection. The bestperforming algorithm, with a classification accuracy rate of 94.9% achieved by performing feature selection on an imbalanced dataset, was Sequential Minimal Optimization. Furthermore, the findings reveal that the majority of users' satisfaction with earthquake applications is relatively high and the overall user experience exhibits a positive trend.