Collaborative Filtering Recommendation Algorithm Based On Improved K-means Algorithm With Central Aggregation Parameters

Nowadays,with the rapid development of the Internet,the information services that accompanying it are also developing rapidly,but there are some problems behind it.When people use the network,people will leave a lot of data information,and these data are growing very fast,which makes the information in the network to be very redundant,people cannot find the content they want quickly and accurately.The emergence of personalized recommendation system has improved this problem and made it convenient to people when they use the network.Therefore,as the core of recommendation system,the performance of recommendation algorithm is particularly important.However,the traditional collaborative filtering recommendation algorithm will cause problems such as algorithm scalability,data sparsity,and cold start with the expansion of the system's scale and increasing data.If these problems can be improved,the performance of the recommendation algorithm will be better.This paper starts from the two aspects of algorithm scalability and data sparsity in order to make some optimization and improving the recommendation quality.The main work is as follows.First,to solve the problem of algorithm scalability,this paper proposes an collaborative filtering recommendation algorithm based on improved K-means algorithm with central aggregation parameters.This algorithm proposes a kind of central aggregation parameters.The purpose of the central aggregation parameters is to screen out the most suitable and optimal initializing cluster center.The improved k-means clustering algorithm with central aggregation parameters is verified on the UCI data set.The results of Adjusting Rand index,Mutual Information and Fowlkes-Mallows index show that the improved algorithm has the best clustering effect.Then,MovieLens data set is used to implement the collaborative filtering recommendation algorithm based on improved K-means algorithm with central aggregation parameters,cluster the score data,and also cluster the corresponding users.These clustered objects have certain similarity,it can reduce the nearest neighbor space for the target users to search,after obtaining the clusters,the traditional collaborative filtering recommendation algorithm is used to calculate the prediction score;Finally,the Mean Absolute Error(MAE)is used to calculate the accuracy,after three groups of comparison experiments,the algorithm has the smallest MAE,so the recommendation accuracy is the highest.Second,this paper to solve the problem of data sparsity,a user clustering recommendation algorithm based on Slope One is proposed,which is verified on MovieLens data set.This paper by comparing the Root Mean Square Error(RMSE)of the Slope One algorithm,the average user ratings,items scoring average,the global average,items popularity and user activity,select the algorithm with the smallest RMSE value,that is Slope One,and use it to fill the initial score matrix is which has missing values,so zero elements are eliminated,and the sparseness of the data matrix is reduced to a certain extent.Then,the classical K-means is used for user clustering,which reduces the nearest neighbor search space of the target user and makes the matching nearest neighbor more accurate.At last,the traditional collaborative filtering recommendation algorithm is used to predict the user's score and MAE is used to calculate the accuracy,three groups of comparison experiments were also performed,and it was found that the algorithm has the smallest MAE,so the recommendation accuracy is the highest.Through the experimental analysis,it is concluded that the two recommendation algorithms proposed in this paper are superior to the traditional collaborative filtering recommendation algorithm in recommendation accuracy.After two improvements,the Mean Absolute Error is reduced to make the recommendation more accurate,which indicates that the scalability of the algorithm and the data sparsity can be improved effectively.To some extent,it enriches and develops the theoretical achievements of the existing recommendation system.It also has a practical meaning,as the amount of data on the network increases and there is little scoring information,the algorithm in this paper can effectively reduce the time to find the similar user range,ensure the recommendation quality is good,and help users get recommendations that better fit their needs,which has a better recommendation effect in the user-based recommendation method.