| With the rapid development of Internet technology,people have gradually moved from the era of information deficiency to information overload,which makes enormous challenges for producers and consumers of information.As an information filtering technology,recommender systems have been widely applied in e-commerce and video fields.It helps users quickly and accurately search for interesting content to meet their individual needs.However,the performance of recommendation is easy to be affected when facing data sparsity and cold-start issues.Therefore,the research on collaborative filtering recommendation algorithm for sparse data has important theoretical significance and commercial value.This paper firstly focuses on the problem of sparse data,and accordingly proposes a new item based collaborative filtering algorithm to improve the accuracy of prediction in sparse data.Then,aiming at the trade-off between accuracy and efficiency,a new collaborative filtering algorithm combining similarity and pre-filtering mode is presented.The main contributions are described as follows:(1)Aiming at the quality of the neighbors and the ability of accurate predictions are affected by sparse data,an item similarity and rating prediction algorithm for sparse data is proposed.First,based on the association between fuzzy sets and recommender systems,user preference probability is calculated and extended to the Vague set,and the KL divergence based on the Vague set is presented to measure item similarity.Then,in order to avoid the loss of rating quantity information,a weight factor is defined and added to the KL divergence to further improve the accuracy of similarity calculation.Finally,an item-based prediction method with a new neighbor selection strategy is presented,which relaxes the strict requirements on the nearest neighbors in the traditional prediction method,and improve the prediction accuracy.Experiments on datasets with different sparsity show that the proposed algorithm has high prediction and recommendation quality,and effectively alleviates data sparsity problem.(2)In order to better balance the relationship between recommendation accuracy and time efficiency,a collaborative filtering algorithm combining similarity measure and prefiltering mode is proposed.Firstly,in user-based hybrid similarity measure,the relative rating difference and the qualitative conditions that should be satisfied is defined to optimize the similarity results.Meanwhile,the improved rating preference based on information entropy and the user’s global rating quantity information are considered as weighting factors,in order to distinguish differences between users and improve the reliability of similarity calculation in sparse data.Secondly,a pre-filtering mode is proposed based on the implicit constraints in the similarity and prediction formulas.Under the condition of keeping the recommendation results,the users and the corresponding rating that do not participate in the calculation,are filtered out to further improve the calculation efficiency.Finally,experimental results on three public datasets show that the proposed collaborative filtering algorithm that has good prediction and recommendation quality while maintaining a low time cost.There are a total of 24 figures,14 tables and 83 references. |
