A Personalized Recommendation Algorithm Based On Collaborative Filtering

In today’s era of rapid development of network information technology,the amount of information in the network is showing a pattern of rapid growth,and the rapid growth of data information has seriously affected people’s judgments,resulting in the inability to obtain the information content that users want to know in a timely and efficient manner.In this era,the personalized recommendation system was developed in line with the trend of the times.According to the research,the collaborative filtering algorithm is the most used algorithm in the current recommendation system.This thesis focuses on the personalized recommendation algorithm based on collaborative filtering,and improves it,so that the algorithm can more effectively meet the needs of users.Firstly,the related recommendation technologies and recommendation algorithms that are widely used in the recommendation system are introduced,the advantages and disadvantages of the recommendation algorithms are introduced,and the shortcomings of the algorithms are improved.Finally,through experimental verification,it is concluded that the improved algorithm can effectively improve the recommendation quality of the recommendation system.The research content of this thesis is mainly composed of the following parts:1.By studying the traditional user-based collaborative filtering algorithm,it is found that the popular objects in the recommender system will not be considered when calculating the user similarity,which will have a certain impact on the calculation of the user similarity.Influence,this thesis adds a penalty factor that can suppress popular objects when calculating user similarity,thereby improving the accuracy of recommendation results.2.Another disadvantage of recommender systems is the problem of data sparsity.In order to perform well personalized recommendation,user-item rating data is very important for calculating user similarity.However,there are very few rating data available in reality.In order to solve the problem of data sparsity,this thesis proposes to use an improved K-Means clustering algorithm to fill the user-item rating matrix.After studying the traditional K-Means clustering algorithm,it is found that the algorithm can produce different clustering results from different initial clustering centers,and the accuracy will be very different.In order to deal with the problem of inaccurate clustering results caused by the random selection of initial clustering centers,this subject proposes to improve the K-Means clustering algorithm based on Kruskal algorithm,so that it can effectively cluster items or users,respectively.Solve the data sparsity problem.3.There are certain defects in a single recommendation algorithm.In order to improve the quality of recommendation that users care about most,this thesis proposes a new algorithm that can solve this problem,that is,a hybrid recommendation algorithm.The hybrid algorithm can fully reflect the advantages of the content filtering algorithm and the collaborative filtering algorithm.To this end,the collaborative filtering recommendation algorithm and the content filtering recommendation algorithm are effectively weighted and integrated,and finally the final recommendation result is obtained by adjusting the weighting coefficient.Improve recommendation accuracy.