Research On Improved Collaborative Filtering Recommendation Algorithm

With the increasing development of computer technology,the amount of data and information in the network shows an explosive increase,and the whole society is entering the era of information overload,and people urgently need efficient information filtering methods.Personalized recommendation algorithm is one of the effective means to deal with information overload,which can directly recommend the information and products that users may be interested in to users.Collaborative filtering recommendation algorithm is the most mainstream personalized recommendation method,which has provided personalized recommendation for users by exploring their preferences on the existing users’ historical data,and has been widely used.However,the current collaborative filtering recommendation methods also have shortcomings,such as data sparsity,lack of scoring timeliness,inaccurate similarity calculation,cold start,and difficulty in feature extraction.These problems seriously affect the effect of the recommendations.For the above problems,this paper carries out the research of collaborative filtering recommendation algorithm based on hybrid model.The main works include:(1)For the problems of data sparsity,lack of scoring timeliness and inaccurate similarity calculation,a co-filtering recommendation algorithm based on improved K-means and optimized scoring is proposed.First,the Weigh Slope One algorithm is used to predict and populate the null value in the scoring data;then improve the K-means clustering algorithm to cluster the users,calculate the user similarity in the cluster,and introduce time weights to improve the recommendation timeliness and similarity calculation accuracy.Experimental results show that the proposed algorithm effectively improves the recommendation accuracy.(2)For the inaccurate problems of data sparsity,cold start and scoring prediction,a collaborative filtering recommendation algorithm based on LSTM emotion analysis and PMF matrix decomposition is proposed.Firstly,the LSTM emotion analysis model is established,which adds the prediction score from the user comments to the original scoring data,then combines the Weight Slope One algorithm to optimize the scoring matrix,and uses the implicit user and item scores to predict the user score.The experimental results show that the recommendation effect of the proposed algorithm improves significantly.(3)We propose a collaborative filtering recommendation algorithm based on probability matrix factorization and variational autoencoder.Firstly,the variational autoencoder is improved to reconstruct the scoring data and then the probability matrix decomposition predict the implicit scoring features of users and items.Experimental results show that the proposed algorithm effectively improves the recommendation effect.In summary,this paper proposes three solutions for the problems existing in the current recommendation system,namely,collaborative filtering recommendation algorithm based on improved K-means and optimization score,collaborative filtering recommendation algorithm based on LSTM emotion analysis and PMF matrix decomposition,collaborative filtering recommendation algorithm based on probability matrix decomposition and variational autoencoder,which effectively reduces the impact of data sparsity,lack of scoring timeliness,inaccurate similarity calculation,cold start,and difficult feature extraction.