Jiang and his colleagues proposed a probabilistic
factor analysis framework, which fuses
users’ preference and social influence together.
Furthermore, they have also investigated the
social recommendation problem in a multiple
domain setting. Most of these works are based
on traditional content-based filtering or CFbased
methods, and their common goal is to
embed social information into traditionalmethods
to improve the recommendation accuracy.
However, few authors have targeted the problem
of how to learn a new common representation
for users and items in social networks, which is
indeed feasible and important for boosting
social recommendation performance.