A similar idea was proposed byWetzker et al. [41], where the probabilistic latent semantic analysis (PLSA) model [10] is extended with tags for the recommendation of resources. In the standard PLSA, the probability that a resource co-occurs with a given user can be computed by
--
where Z := {z1, ..., zq} is a hidden topic variable a--nd is assumed to be the origin of observed co-occurrence distributions between users and resources. The same hidden topics are then assumed to be the origin of resource/tag co-occurrences, i.e.,