Diversity between Human Behaviors and Metadata Analysis: A Measurement of Mobile App Recommendation

Abstract

The explosive growth of mobile apps has given rise to the significant challenge of app discovery. To meet this challenge, the Google Play market utilizes the user behaviors data to provide app recommendations. By making use of experiences of the user crowd, such recommendations are of help to users for discovering apps. However, they are concurrently restricted to the local scope of the user experiences, as most users have only accessed a limited amount of apps. To conquer this constraint, we propose a novel recommending method by utilizing the global information of apps. To be specific, we leverage the Latent Semantic Indexing method to analyze the metadata of apps, which is globally held by the market. We thus obtain the similarity measurements among apps and based on them we generate app recommendations. To further understand both the human behavior based and the metadata analysis based methods, we then measure the diversity within them from multiple levels and scopes. Through such measurements, we eventually discover new knowledge of user preferences and gain better understanding of both recommending methods. These observations further indicate that there are necessities and potentials to evolve the existing mobile app recommender systems by integrating new recommending methods.