Research On Collaborative Filtering Recommender System Based On Data Feature

With the rapid development of big data technology and artificial intelligence,recommender system has been everywhere in people's lives in recent years.Personalized recommendation has also attracted a lot of attention from both academia and industry.Many excellent research results have been reported.This paper analyzes the hidden data feature in similarity matrix of different recommendation algorithms based on principle component analysis(PCA)and uses it to improve the recommendation algorithm.The detailed research contents are as follows:(1)An improved collaborative filtering recommendation algorithm based on data feature research.Firstly,this paper finds that most eigenvectors contain little information,and the eigenvectors corresponding to the maximal eigenvalue contain most information by analyzing the contribution rate of the eigenvectors of the similarity matrix and the distribution of the eigenvalues.Secondly,this paper studies the relationship between the absolute value of the eigenvector corresponding to the maximal eigenvalue and the object popularity in recommendation list.The result is positive correlation.As a result,a general function form of improved algorithms is proposed and used to improve different recommendation algorithms.The experimental results based on four recommendation datasets show that the improved algorithms have better recommendation performance and can effectively promote the recommendation accuracy of the unpopular objects.Moreover,this paper finds that I?HHM algorithm and I?BHC algorithm with the best recommendation after improvement can still maintain good recommendation accuracy when the rating information of historical users is reduced.This illustrates that the algorithms can effectively alleviate the negative impact of sparse historical rating information on the system and have certain reliability.The general form of the improved algorithm proposed in this paper is parameter independent and has advantages in practical application of recommendation systems.(2)An improved collaborative filtering recommendation algorithm by optimizing the data-feature-based similarity.Since the selection of datasets and recommendation algorithms would affect the accuracy,diversity and novelty of the recommendation results,this paper proposes an optimization algorithm based on the framework of the improved recommendation algorithm.There are two adjustable parameters in the functional form of the optimized P?HHM algorithm and P?BHC algorithm.Therefore,this paper takes a method of optimizing another parameter based on the optimal parameters of the algorithm itself.At last,the experimental results on the four recommendation datasets show that the accuracy,diversity and novelty of optimized recommendation algorithms are further improved.The optimized algorithms make the popular objects and unpopular objects in the recommendation list more uniformly distributed and better balance the accuracy and diversity.It can not only exactly predict the preferences of users,but also effectively reduce the situation that users are tired of similar recommendations.To sum up,this paper extracts the data feature embodied in algorithm and discovers the relationship between the absolute value of the eigenvector corresponding to the maximal eigenvalue and the popularity of objects,which has been used to improve and optimize the collaborative filtering recommendation algorithm.Experimental results show that the improved and optimized algorithms can enhance the recommendation performance effectively.