A Hybrid Collaborative Filtering Algorithm Based on KNN and Gradient Boosting

Abstract

The recommendation systems are widely used in e-commerce, especially for the recommendation systems based on collaborative filtering (CF) methods. It can analyze users' interests and preferences from their historical data, and then recommend items to them. In existing methods, it is common to calculate the similarity between users, and predict the users' rating for the items. However, most of them are predicting the ratings by single classifier and the effect is not very good. In this paper, we proposed a hybrid CF method combined the k-nearest neighbor (KNN) algorithm and gradient boosting method. We consider the effect to the performance both of the similar users and similar goods. We calculate the correlation between users and items with the KNN algorithm to filter out similar users and similar items, and then we use the gradient boosting with ensemble learning to predict users' ratings for the items. Compared with existing methods, we consider the information about similar users and similar items, and adopt multi-classifiers with ensemble learning to predict the items that users may like. We can achieve better effects in the performance. Therefore, we proposed this method to recommend items for users.