Feasible settings for the adaptive latent semantic analysis: Hk-LSA model

Recently, improvements of latent semantic analysis or LSA which stems from singular value decomposition to derive latent semantic classes, especially hk-LSA model, have been proposed. The hk-LSA model is based on reducing dimension of vector space and like-probabilistic relationship between document-term and latent-topic space. This improved model overcomes some shortcomings of standard LSA such as processing very dense and orthogonal matrices and difficulties in parallelization. It is dealt with this paper, some feasible ways to setup such a model and statistical comparisons between proposed ways to recognize good setup feasible for the hk-LSA model. Case studies on this subject suggest some ways to setup hk-LSA and show relationships between the standard LSA and hk-LSA model.