Personalized Recommendation Algorithm Research Based On Incomplete Information Feature Extraction And Hidden Markov Model

As the rapid development of Internet technology brings in the overload information,the personalized recommendation system emerges.However the traditional personalized recommendation system owns single,static,overlapping attribute,strong dependence and other features,as a result the recommendation is not desirable.Besides the scale of information expands continuously,incomplete information appears.Then it is not available to achieve precise data analysis and the accuracy of recommendation algorithm is decreasing.At the same time the traditional recommendation algorithm has the problem of data sparseness,extensibility and low recommendation accuracy.Thus effective feature extraction from the incomplete data information and solution to the data sparsity for recommendation accuracy are to solve for current recommendation algorithms.This paper studies the problems of the traditional recommendation algorithm,which mainly focus on incomplete information feature extraction as well as personalized recommendation algorithm based on Hidden Markov Model by fusion of explicit and implicit user behavior feature.Finally this paper verifies the related research methods by experiments.And the paper mainly finish the following task:(1)Research effective feature extraction methods from the incomplete data information.For the incomplete data information,we can obtain valuable rules by feature extraction from the incomplete data information to promote the accuracy of data analysis.This paper presents a large-scale incomplete information feature extraction algorithm.The algorithm utilizes neural network to describe the relationship between incomplete information attributes and classification results,and uses the derivative relationship to build a classification decision tree,and clusters incomplete information.Then using adaptive search method to obtain the center vector value of feature attribute category,thus achieve the fuzzy control law of information feature extraction under dynamic training,calculate feature mining fitness value,and realize large-scale incompleteness.(2)Research personalized recommendation algorithm based on Hidden Markov Model by fusion of explicit and implicit user behavior feature.Aiming at the problem of data sparseness,extensibility and low recommendation accuracy of the traditional recommendation algorithm,this paper proposes a recommendation algorithm based on Hidden Markov Model by fusion of explicit and implicit user behavior feature.Firstly the algorithm obtains implicit user behavior features(clicks,page remaining time)and implicit user behavior(scores)of Chapter 3,then fuses the three behavior features into 1 to 5,and then performs users,user behavior(merging scores)and projects to hidden Markov modeling and learning to acquire the optimal Hidden Markov model.And according to forward-backward algorithm,this recommendation system computes scoring probability for certain items at the next moment and the Viterbi scheme is used to estimate the item scores.Finally,these highscoring and high-probability items are recommended to the user.(3)Experimental analysis was performed on the movie data set.This paper proposes a recommendation algorithm based on Hidden Markov Model by fusion of explicit and implicit user behavior feature.The results show that compared with the traditional recommendation algorithm,the proposed algorithm has significant promotion in accuracy,recall rate and F1 index.