A Combined Predictor for Item-Based Collaborative Filtering

Abstract

Collaborative filtering is one of most important technologies in the field of recommender systems, the process of making predictions about user preferences for products or services by learning known user-item relationships. In this paper, slope one and item-based nearest neighbor collaborative filtering algorithms are analyzed on the Movie Lens dataset. In order to obtain better accuracy and rationality, a new combined approach is proposed that takes advantages of slope one and item-based nearest neighbor model. In addition, simple gradient descent and bias effects are used further to improve performance. Finally, some experiments are implemented on the dataset, and the experimental results show that the proposed final solution achieves great improvement of prediction accuracy when compared to the method of using slope one or item-based nearest neighbor model alone.