matrix factorization from a small number of observed entries has recently
garnered much attention as the key ingredient of successful recommendation
systems. One unresolved problem in this area is how to adapt current methods to
handle changing user preferences over time. Recent propos