| Abstract |
Mobile application stores have recently gained significant popularity due to the evolution of the smartphone applications' (apps) market and the large increase of smartphone users. Apart from official marketplaces such as Google's Android Market or Apple's App Store , there is a plethora of alternative app stores with a large number of both applications and users.
In this thesis, we perform a systematic study on four third-party Android marketplaces, in order to improve our understanding on several aspects of this rapidly evolving ecosystem. More specifically, we study (i) how apps are being produced and common strategies observed among app developers, (ii) the app popularity pattern, (iii) user download patterns and how can be affected by recommendation systems, as well as (iv) how app pricing affects apps popularity and the developers' income. Furthermore, we compare our findings with similar studies on other fields, such as the world wide web (WWW), peer-to-peer file sharing (P2P) systems, and user-generated content (UGC) sites.
The data of our analysis are gathered by systematically crawling four popular Android appstores in a daily basis, for several months. Our results indicate that mobile marketplaces are comprised of a small number of very popular applications that absorb the majority of downloads, confirming the existence of the Pareto Principle. Moreover, the distribution of app popularity follows a Zipf-like behavior with some deviations, very similar with the behavior observed in P2P and UGC workloads. We verify that these deviations from the Zipf distribution are caused in part due to the "fetch-at-most-once" user behavior, that other studies have already demonstrated, as well as by the existence of another phenomenon we call the clustering effect. According to the clustering effect, apps are grouped into clusters, which may be a result of recommendation systems, user communities or other grouping forces, and users tend to download apps from the same clusters with higher probability. We verify our app clustering hypothesis using a new metric called "user temporal affinity" to app categories, and we measure it using a dataset with user comments, which implies user downloads. The results show that indeed users have a strong affinity to app categories. Then, we propose a novel model based on the clustering effect and fetch-at-most-once property, and we evaluate our model with a simulation-based study comparing with the observed app downloads. We find that our model approximates very well the actual distribution of app downloads.
|
Rise of the Planet of the Apps: A Systematic Study of the Mobile App Ecosystem
