Social influence modeling on smartphone usage

Abstract

This paper presents a probabilistic influence model for smartphone usage; it applies a latent group model to social influence. The probabilistic model is built on the assumption that a time series of students' application downloads and activations can be represented by individual inter-personal influence factors which consist of latent groups. To verify that model with its assumption, about 160 university students voluntarily participated in a mobile application usage monitoring experiment. Analysis could identify latent user groups by observing predictive performance against reduced dimensions of factor matrices with NMF. Proper dimension reduction is shown to significantly improve predictive performance, which implies a reduction in the over-fitting phenomenon. With this improvement, the model outperforms conventional collaborative filtering models and popularity models in perplexity evaluation. The results validate the model and its assumption as well as its usefulness. © 2011 Springer-Verlag.