A Joint Model for Who-to-Follow and What-to-View Recommendations on Behance.

Good recommendations are a key tool to increase user engagement and user satisfaction on many social networks. Here we focus on Behance, a social network for artist from various fields such as typography, street art, industrial design, and fashion. On Behance, the artists can connect by following each other, display their work in online portfolios, and brows each other's work. Each user has a personalized dashboard which is an integral part of the Behance experience. In this work we create a joint behavior model which jointly models the users' viewing behavior and the social network. The joint model which we fit with variational inference is capable of producing both who-to-follow and what-to-view recommendations. We show on real data from Behance that the joint behavior model outperforms a Poisson factorization approach which treats both data sources separately.