Collaborative filtering using graph kernel and boosting

摘要

Based on the users previously given preferences, recommender systems intent to find similar things and give predictions to them. One of the most popular methods for recommendation is Collaborative filtering and it is the workhouse of the recommender engines. A lot of conventional recommender systems result in failure when the data to be used all over the recommendation process is sparse. So inorder to overcome the problem of sparseness of data, graph based methods can be introduced, since they targets situation where labeled data are scarce and unlabeled data are in abundance. In this work graph based methods like graph kernel and boosting are taken as a framework for the collaborative filtering inorder to get promising results. Here as a boosting method, most popular AdaBoost algorithm is chosen and from the graph kernels, regularized Laplacian kernel is used. Both these methods are compared with the normal collaborative filtering algorithm using the same dataset and the results shows that the system is more accurate when using the graph based methods.