Measuring and Understanding Crowdturfing in the App Store

Abstract

Application marketplaces collect ratings and reviews from users to provide references for other consumers. Many crowdturfing activities abuse user reviews to manipulate the reputation of an app and mislead other consumers. To understand and improve the ecosystem of reviews in the app market, we investigate the existence of crowdturfing based on the App Store. This paper reports a measurement study of crowdturfing and its reviews in the App Store. We use a sliding window to obtain the relationship graph between users and the community detection method to binary classify the detected communities. Then, we measure and analyze the crowdturfing obtained from the classification and compare them with genuine users. We analyze several features of crowdturfing, such as ratings, sentiment scores, text similarity, and common words. We also investigate which apps crowdturfing often appears in and reveal their role in app ranking. These insights are used as features in machine learning models, and the results show that they can effectively train classifiers and detect crowdturfing reviews with an accuracy of up to 98.13%. This study reveals malicious crowdfunding practices in the App Store and helps to strengthen the review security of app marketplaces.