Modeling Users' Adoption Behaviors with Social Selection and Influence.

Ziqi Liu,Fei Wang,Qinghua Zheng
DOI: https://doi.org/10.1137/1.9781611974010.29
2015-01-01
Abstract:Massive users’ online adoption behaviors were recorded thanks to the various emerging web services such as Facebook, Twitter, G+, Netflix and so on. Two key factors that affect users’ adoption behaviors are social selection and social influence. Understanding such factors underlying each behavior can potentially help web service providers gain much more insights into their users and improve predictive power. In this paper, we try to answer (1) How do the roles of selection and influence play in a user-level adoption? (2) Capturing those factors can benefit the modeling and prediction of users’ adoption behaviors or not. Quantitatively capturing the two factors could be challenging since the known ”ballot box communication”. Moreover, though both social selection and influence are well studied in collaborative filtering and information diffusions respectively, it’s still non-trivial to jointly model them. We propose a probabilistic Latent Factors with Diffusion Model (LFDM) which explicitly considers both social selection and influence by projecting cascading processes into latent factor spaces. We also develop an effective EM styled algorithm for estimating the proposed model. Finally we validate our methodology on three kinds of real world data sets.
What problem does this paper attempt to address?