Detect Similar Mobile Applications with Transfer Learning

Abstract

Recent years have witnessed the fast growth of the use of the mobile applications (a.k.a. apps). Detecting similar apps is a basic problem in the app ecosystem. It is not only beneficial to app search and recommender systems, but also helpful for people to discover new apps. State-of-the-art studies defined several app similarity functions by the metainformation of apps, such as descriptions and reviews, and succeeded to apply it to apps from Google Play. There do exist some app stores that provide similar app lists when people search. Obviously such similarity relation information will help improve similar app recommendation in app stores where app recommendation is of poor quality or even does not exist. However few methods take the existing similarity relations between apps from different app stores into consideration. Our approach attempts to use such relation information to learn a new app similarity function from these stores, and combine this new function with existing functions defined from meta-information of apps. With transfer learning, our proposed similarity function can be used in app stores where no similar apps are presented. Empirical results on Chinese app data sets show that our method outperforms the state-of-the-art method significantly.