Detecting Fake Apps Through User Review Analysis on the Google Play Store

Abstract

As mobile applications rapidly grow in popularity on platforms like the Google Play Store, users face an increasing threat from fake or malicious apps. Numerous applications take advantage of user trust by mimicking well-known apps or by manipulating the review system to seem more trustworthy. This research presents a deep learning-based approach to detect fake apps by examining user reviews using advanced Natural Language Processing (NLP) techniques. The review data was gathered from various apps on the Google Play Store and underwent necessary preprocessing. Sentiment and contextual clues were extracted utilizing transformerbased models. Three prominent pre-trained models, RoBERTa, DistilBERT, and BART-large, were fine-tuned to categorize apps as authentic or fake based on their reviews. These models were assessed with metrics such as accuracy, loss, ROC curve, and precision-recall scores. RoBERTa attained the highest accuracy at 99.12%, followed by DistilBERT at 98.42% and BART-large at 97.96%. A comprehensive system that integrates web scraping, model inference, and a user-friendly interface was developed. This platform enables users to enter an app link and receive instant predictions regarding its authenticity.