Predicting personalized grouping and consumption: A collaborative evolution model

Abstract With the prevalence of online social groups, the dynamic joint prediction of users’ grouping and consumption behaviors on social network platforms is critical for optimizing social link suggestions and product recommendations. The group influence theory indicates that group norms affect user preference and behavior; however, the individual preferences of group members can also alter group norms. Nevertheless, the problems of how to holistically model dynamic bidirectional influence, existing between individual preferences and group norms, and then, simultaneously predict the users’ grouping and consumption behaviors are still underexplored. In this study, we propose a collaborative evolution and prediction (CEP) model to address the above issues. We associate each social group with a latent group norm vector, and assign each user with a latent individual preference vector. The unobservable interplay between individual preferences and group norms is then modeled according to the underlying group influence theory. Based on these two latent vectors, we design a joint optimization function that incorporates the correlation between grouping and consumption behaviors, to enhance the prediction performance. Through extensive experiments and evolution analysis, we demonstrate the prediction effectiveness and the explanatory power of our CEP model.