Gizlilik tabanlı ortak filtreleme sistemlerinde kayıp değerlemeleri işleme

Collaborative filtering is an influential personalized recommendation technique deducing like-minded users from their ratings and producing predictions for them. However, the first controversial issue with this technique is that people may share a lot of individual information with collaborative filtering systems, which brings serious privacy risks. Privacy-preserving collaborative filtering algorithms are mainly contrived to deal with this privacy challenge. Missing values in the collected data set is another major issue in collaborative filtering systems. Users usually do not rate all items; conversely, they rate only a limited number of them because there are too many items to rate. Accordingly, there exists insufficient information to locate similar users correctly and generate accurate predictions. There are readily available methods in the literature constituted to overcome this problem. While some of these methods try to impute the missing values by only using the available data, the others utilize auxiliary data. The objective of this study is to apply some of the missing data imputation methods using no auxiliary data on several privacy-preserving collaborative filtering algorithms in order to boost the recommendation quality. Existing missing data imputation methods are modified in such a way that they can be applied to perturbed data. Several experiments are performed using a real data set to show how effective the methods are in privacy-preserving collaborative filtering systems.