Genetic algorithm based privacy preserving collaborative filtering

With the increase of internet usage, people started to meet almost every need from online service companies. However, as the number of users and products and the variety of products offered by the companies providing services over the internet increased, it became increasingly difficult for users to decide on product selection. Recommendation systems, by automating these processes, offers users the most suitable product at the decision-making stage. Recommendation systems should be able to produce correct recommendations safely within an acceptable period of time. In this thesis, it is aimed to meet all of these criteria. It is intended to produce more accurate suggestions to the users by improving the closest neighborhood based collaborative filtering technique which is a memory based traditional technique. On the other hand, the data security of the system has been ensured by preserving the privacy of the rating information given by the users to the products. Collaborative filtering and privacy-preserved collaborative filtering methods were developed with separate genetic algorithm and the results were analyzed. In both methods, it was observed that the system produced more accurate results than traditional techniques. It is observed that the increase in the overall accuracy of the privacy-preserved collaborative filtering method after it has been improved by genetic algorithms is proportionally better than the improvement obtained in the collaborative filtering method that is not private and the reasons for this manner are explained.