Weighted Slope One Algorithm Optimization Based On User Similarity And Item Similarity

According to the characteristics of different users' interests,personalized recommendation technology makes targeted recommendations.The core of this technology lies in the design of the recommendation algorithm.The better the algorithm design is,the better the recommendation effect will be.At present,recommendation algorithms can be divided into content-based filtering algorithm(conf-algorithm),collaborative filtering algorithm(colf-algorithm)and hybrid recommendation algorithm(HR-algorithm),among which the collaborative filtering algorithm can be subdivided into memory-based collaborative filtering algorithm and model-based collaborative filtering algorithm.The weighted Slope One algorithm optimized in this topic belongs to the memory-based collaborative filtering algorithm.As a collaborative filtering algorithm based on memory,weighted Slope One algorithm has the greatest advantage of simple principle,easy implementation,high execution efficiency and relatively high accuracy in predicting score.Moreover,it supports online query and dynamic update,which makes it an excellent candidate for recommendation algorithm in the real world.Different from the traditional collaborative filtering algorithm,weighted Slope One algorithm does not calculate the similarity between items or users,but uses a simple linear regression model to predict the score,and uses the data of all users and projects without distinction,which is likely to cause deviation in the prediction of the score of target items,thus affecting the recommendation quality.In view of the problem that weighted Slope One algorithm fails to fully consider the internal correlation between users and between projects,this topic proposes an optimization strategy of weighted Slope One algorithm based on user similarity and project similarity.The main work is as follows:1.As for the weighted Slope One algorithm,it does not take into account the possible correlation between different users and target users when calculating the mean of inter-item scoring deviation.In this study,the similarity among users is introduced into the original algorithm formula as a weight factor.On user similarity calculation,this paper puts forward an improved strategy of Pearson correlation coefficient using Pearson coefficient and the normalized Euclidean distance linear combination way of make up for the inadequacy of Pearson correlation coefficient to calculate the user similarity,at the same time in the choice of nearest neighbor set to target users,put forward a kind of similarity degree of support screening strategy.2.For the weighted Slope One algorithm does not consider the correlation between the items that have been graded by target users and target items,it proposes to introduce item similarity as a parameter into the original algorithm.A new item similarity calculation method is designed and integrated into the original calculation formula,which reflects the similarity between items as a whole by integrating item type label similarity and item score similarity.The improved algorithm added user similarity and item similarity as weighting factors to the original formula,which improved the accuracy of scoring prediction and increased the computational complexity at the same time.The experimental analysis of Movie Lens data set shows that the optimized algorithm improves the accuracy of 3%-4% on the basis of the original algorithm,and achieves better prediction accuracy compared with other two optimization methods for weighted Slope One algorithm.