Data integration in recommendation systems

The main goal of recommendation systems is to predict the user pleasure and to recommend something in the boundry of its pleasure. In recommendation systems, the most effective techniques are ;?Content - Based Filtering?Collaborative Filtering?Hybrid Filtering Systems?Demographic Filtering?Knowledge - Based FilteringContent - Based Filtering is mainly interested about the similarities of content between the current item (music, movie, book, etc.) and the others. In opposition of Content - Based, in Collaborative Filtering system is interested in interaction of users. The choice of the current user depends on the choices and prevision of the other users or user groups. According to prevision of users, the system recommends similar item to the current user.Data Integration is a method that uses users evaluation vectors on items. The aim of this operation to get more complex vectors which have more features by joining the horizontal and vertical vectors of datasets.In this thesis, a new hybrid system is proposed. On the prediction values which are produced by content based filtering using by metadata of items; "Rating of one user on all items" information and "Ratings of all users on one item" information is integrated. After this integration; "Support Vector Machine" and "K - Nearest Neighbor?, the most popular classification algorithms, are executed on these new datasets. The main goal of this thesis is to increase the forecast perfomance of recommendation systems by data integration.Finally in the comparison of this technique and other techniques shows us that the data integration technique gives better results according to F- measure. On the other hand, there is also a comparison according to resource usage. "Support Vector Machine" (SVM) uses more memory and takes more time to solve the problem. However "K - Nearest Neighbor? (KNN) uses much more CPU.