Mobile Application Recommendation based on Demographic and Device Information

Abstract

The number of people with access to mobile devices, as well as applications to these devices (i.e., apps), has been increasing significantly. Thus, users have to choose among a high number of apps, those that better serve them. However, this is not a trivial task, as we have seen an increasing number of apps proposing to do the same functions. In the same way, companies are facing difficulties to attract users through a usual marketing campaign. A possible solution to this problem is the adoption of recommendation systems, where it is possible to compare the similarities of user profiles. Meanwhile, these systems often consider only users' preferences to create a profile, or request sensitive data (e.g., call and message logs). However, the installation of an app may involve other factors like the capacity of the mobile device (e.g., memory and processing power) and the demographic information of the user's living area. This work investigates the impact of using demographic and handset information on app recommendation. To do that, we use this information to enrich a user profile that has only easy-to-obtain data (i.e., installed apps, approximate location, and handset model). Besides, our proposal was evaluated on three different recommending approaches: Latent Dirichlet Allocation (LDA), Markov Transition Matrix (MTM), and Collaborative Filtering. The general results reveal that the LDA approach achieved the highest efficacy when added information about the user's region mean wage, in terms of precision (approximately 64%) and recall (approximately 28%).