Determining the best recommendation algorithms for user/item pairs

With the development and the spread of the Internet and communication technologies, the services offered via the Internet has increased. As a result of this increase, users are faced with thousands of products and information to be examined and followed. This situation is defined as "Information Overload". Recommenders systems have been developed to assist in the selection of products and services to the user. The most popular recommendation system used to provide personalized recommendations based on users' preferences for the past and similar users in online services is the Collaborative Filtering based recommendation systems. Accuracy is one of the most important features that must be satisfied by the Collaborative Filtering based recommendation systems. The aim of this thesis is to improve the accuracy of the recommendation produced by the Collaborative Filtering based recommendation systems. Today Collaborative Filtering Recommender systems typically utilizes only one recommendation algorithm and provide prdictions for all users by using the same algorithm. The hypothesis is that, there is one best algorithm for each user or user/item pairs, instead of hiring the same recommendation algorithms for all users or user/item pairs, hiring the best recommendation algorithms for different users or user/item pairs can increase accuracy of recommendation system. As a result of experimental study it is proved that, there is one best algorithm for each user or user/item pairs, hiring the best algorithm can increase accuracy of Collaborative Filtering recommendation system.