A Tutorial on Probabilistic Latent Semantic Analysis

Historically, many believe that these three papers [7, 8, 9] established the techniques of Probabilistic Latent Semantic Analysis or PLSA for short. However, there also exists one variant of the model in [11] and indeed all these models were originally discussed in an earlier technical report [10]. In [2], the authors extended MLE-style estimation of PLSA to MAP-style estimations. A hierarchical extension was proposed in [6]. In [4], the authors showed the equivalent between PLSA and another popular method, non-negative matrix factorization. A high order of proof was shown in [12]. The equivalent between PLSA and LDA was shown in [5]. More recently, a new MAP estimation algorithm is proposed in [13].