The Research Of Probability Latent Semantic Analysis In Multidimensional Recommendation System

With the rapid development of Internet technology,a wide range of information transfer in the network.The network allows people to enjoy the entertainment,access to knowledge,but the information screening is a question we have to face.How to provide a suitable information for customers in an efficient and accurate manner,that is very urgent in today's information services.In the real life,we find that the user's scenario environment will change the user's interest in the project preferences,the traditional recommendation system is mainly for the "user-project"two-dimensional recommendation,recommended algorithm that consider the scenario factor is still relatively rare,in order to Get the efficient and accurate recommendation effect,this paper mainly make the following research:(1)Some recommended methods commonly used in the field of two-dimensional recommendation system have been studied,and the algorithm flow of collaborative filtering in multidimensional mode considering the situation information have been studied.Then,the process of SVD decomposition in two-dimensional mode is analyzed.Finally,the algorithm flow of HOSVD tensor decomposition considering scenario information have been studied.(2)First i study the initial application area of PLSA algorithm-"text retrieval".The paper puts forward the idea of applying the PLSA algorithm to the "user-project" two-dimensional recommendation field,and gives the corresponding table of "user-project,and "document-vocabulary" model,i analysis and prove the PLSA algorithm seriously,i extend and improve the PLSA algorithm,then i get the CPLSA algorithm which is used in the "user-project-scenario" Multi-dimensional recommendation system.(3)First i study The evaluation of the effectiveness of the recommended system.Then i make a comparative experiment.Finally,i get the conclusion that the Probability algorithm of PLSA and CPLSA is better than traditional SVD matrix.TheCPLSA algorithm considering the scenario information is better than the two-dimensional PLSA during the recommendation.The number of implied factors k and the number of iterations Q will have an effect on the recommended effect of the PLSA and CPL5A algorithms.