Empirical Study On Underlying Network Structure And Evolution Of Mobile Applications

With the flourish of mobile computing, mobile Apps dominate the daily lives of users. Strategies that help select suited ones from millions of Apps for users are needed. App developers also want to get userful feedbacks from users to help improve their Apps. Plenty of work in the past decades have found there are social relations among Web services, which play a great role in service recommendation field. Mobile Apps have been the dominant service morphology in today’s service-oriented computing paradigm. So, exploring potential social relations among Apps may help imporve App recommendations. As the popularity of Apps increases, maintaining them will become critical. Due to intense competition in the field of App store, developers iteratively upgrade their Apps to improve the quality. A good understanding on the underlying evolution patterns/laws of Apps may be of great benefit. Thus, empirical studies are conducted in this paper:(1) Focusing on the social relations among Apps, we make an empirical study on constructing the Global App Network(GAN) in terms of three types of inter-App relations(i.e., Intent-based, semantics correlation based, and similarity-based ones), recovering Personal App Network(PAN) in terms of App usage log of each user, and exploring the characteristics of GAN and PAN. The study is based on two real-world datasets: the first one includes thousands of Apps collected from a real-world Android App store, and the second one contains 2-month App usage logs of 40 volunteers. Several interesting phenomena are observed from the study, such as 1) a large portion of implicit inter-App relations that are welcome by massive users are actually ignored by App developers; 2) some explicit relations proactively designed by App developers are actually not frequently used by users; 3) although there is a certain commonness among PANs of different users, each PAN shows a significant personalized pattern which delineates the individualized behaviors of a user. These conclusions are of significance to bi-directional App recommendations, i.e., to recommend neglected inter-App relations to App developers, and, to recommend suited Apps to users.(2) Based on the studies of GAN and PANs, we propose two new algorithms on App recommendation, introducing inter-App relations to recommendation process. Compared to traditional collaborative filtering solutions(i.e., matrix factorization), our algorithms show a great improvement both in precison and recall in five different rank positions. We are the first to study the relation recommendation from PANs to GAN and use some cases to certify its value to App developers. In a word, the studies of inter-App relations provide great support for bi-directional App recommendations.(3) We also make an empirical study on App evolution, especially on inter-App interface evolution(e.g., the Intent and Intent-filter of Android Apps) and inner-App feature evolution, both from externally observable exhibitions of Apps. Interfaces are extracted from.APK(Android PacKage) files, and statistical methods are used to discover underlying patterns of interface evolution. Furthermore, potential trend on how interface evolutions of Apps result in the evolution of GAN is observed. Latent Dirichlet Allocation(LDA) is applied to extract updated features(“topics”) from “What’s New” of each version to explore the underlying patterns of feature evolution. We find 1) there are four types of interface evolution patterns, the evolution of Intent is more intensive than the one of Intent-filter, and 2) the density of inter-App relations becomes higher along with time. 3) We identified three types of feature evolution and there is significant “temporal locality” in feature evolution. 4) Case study reveals that there indeed exists more or less feature co-evolution among multiple Apps in their respective evolution. These findings may help developers make strategies of App evolution, reduce maintaining cost and get higher popularity.The empirical study of static social relations among Apps and underlying network provides theoretical basis for the bi-directional App recommendation, which may help users find suited Apps and help developers improve their Apps so as to achieve higher polularity. The empirical study on evolution focuses on the dynamic characteristics of Apps, which may be of great significance to App developers by helping them get a good understanding on the underlying evolution patterns/laws of mobile Apps.