Research On Personalized Recommendation Method Based On Collaborative Filtering Hybrid Filling

With the continuous popularity of the Internet,people have entered the era of information society and network economy.The amount of information available from the Internet is countless,but the effective information is the tip of the iceberg.The recommendation system came into being to link users and information together,and realized the function of mining meaningful information for users from a large amount of data.Collaborative filtering,as the most popular recommendation algorithm,can determine the set of nearest neighbors from users' rating record of items,and analyze the user's preferences from the behavior records of the nearest neighbors.However,as the number of items in the recommendation system continues to increase,the number of items evaluated and viewed by users in a large-scale item collection is very small,resulting in a high sparseness of the recommendation system,which affects the recommendation performance.This paper introduces the relevant knowledge of the recommendation system,explains the recommendation process of the collaborative filtering algorithm,and conducts in-depth research on the problem of low accuracy and low coverage of the collaborative filtering recommendation results due to data sparsity,and proposes two schemes to improve.The main work is as follows:1.This paper proposes a hybrid filling collaborative filtering algorithm(HFCF)to alleviate the problem of data sparseness.First,from the perspective of the item,if users have not rated an item temporarily,the rating that may be evaluated for the item is predicted according to users' ratings of the nearest neighbor items,and filled into the sparse matrix.At the same time,from the user's perspective,use the filled matrix to determine the neighboring users of the target user,select items with the highest number of common evaluations by the neighboring users,and the matrix is further filled according to the neighboring users' ratings on these items.This algorithm does not require other complicated information,which significantly reduces the sparseness of the data.It is experimentally verified that the method effectively improves the accuracy of the recommendation.2.We propose an algorithm based on reduced neighboring user information for filling(RNFCF).Firstly,analyze the attributes contained in the target user's favorite items,select some of the attributes,and items with these attributes will constitute the domain of discourse.Secondly,analyze other users' preference for items in the domain of discourse,all the itemsthat each user likes form a subset,and the collection of subsets constitutes a coverage on the domain of discourse.According to the coverage reduction algorithm,redundant users are removed from the neighboring user set.Finally,according to the rating information of some users with high similarity among the retained neighbor users,the ratings of the target user are predicted,and then filled into the matrix.After filling,recommendations are generated according to collaborative filtering,which effectively avoids the problem of data sparseness.Experiments show that RNFCF effectively improves the coverage and accuracy of recommendations.