| With the rapid development of the Internet and the rapid increase of information,the problem of information overload has arisen.People cannot choose the information they are really interested in from so much information.At this time,the recommendation system was born.It can help people quickly get the expected information when they are faced with huge amounts of data,and it has been applied in many fields.Collaborative filtering algorithm is the most popular recommendation system construction method at present.It is a widely studied technology in the field of data mining and information retrieval and has been successfully applied in many software.In the past few decades,due to the rapid growth of the Internet,a large amount of data has been generated.Collaborative filtering algorithms are faced with the problems of evaluation matrix sparsity,growing data and cold start,which pose challenges to collaborative filtering algorithms.Aiming at the problem of data sparsity in the algorithm,weighted Slope One algorithm is used to fill the vacancy of the scoring matrix in this thesis.Its advantage is that it is simple and easy to implement and can be updated online in time.However,the weighted Slope one algorithm does not calculate the similarity between items,which will cause bias to the score prediction and affect the quality of the recommendation.Therefore,Pearson correlation coefficient with difference factor is used to calculate the similarity of the item type and the item score matrix,and the two are combined linearly,and the final item similarity is obtained by adjusting the parameters,and the similarity is ranked from high to low.The first N items are used as neighbor items,and then the weighted Slope one algorithm is used to predict the score items,and the predicted values are used to populate the original user-item score matrix.The introduction of item type similarity can avoid the interference of items with low similarity and improve the accuracy of the estimated value.Compared with the original weighted Slope One algorithm,the improved algorithm MAE is obviously smaller than the original algorithm.Aiming at the scalability problem of the algorithm,this thesis introduces the idea of clustering.First canopy algorithm was used to perform rough clustering of data,determine the number of clusters,introduce the "maximum and minimum distance product method" to select the initial point to improve accuracy,and then K-means algorithm is used for clustering,after generating multiple clusters,the modified cosine similarity is combined with user attribute characteristics to form a new similarity calculation model,and finally the corresponding recommendation is made.Clustering can reduce the search range of target users and simplify the execution time of the algorithm,so as to improve the scalability of the system.By comparing with the other four collaborative filtering algorithms,it can be seen that the MAE of the algorithm in this thesis is less than the other algorithms in a certain range,which proves that the improved algorithm has higher accuracy,and the experiment shows that the improved algorithm consumes less time.The Movielens data set is used to verify the two algorithms proposed in this thesis,and compared with other algorithms,the experimental results show that the algorithm in this thesis can effectively alleviate the sparsity problem of the score matrix,and has better scalability and other advantages. |
