Collaborative filtering by co-training method

Abstract

Collaborative filtering is a technique to predict the utility of items for a particular user by exploiting the behavior patterns of a group of users with similar preferences. This method has been widely used in e-commerce systems. In this paper, we propose a collaborative filtering method based on co-training– a semisupervised technique that iteratively expands the training set by switching between two different feature sets. In the collaborative filtering settings, our co-training based method uses users and items as two different feature sets. Each feature set is used to infer the most reliable predictions which are then added to the new labeled set. This procedure leads to improved prediction accuracy and reduces the negative influence of data sparsity – a main obstacle to the application of collaborative filtering. The experimental results on real data sets show that the proposed method achieves superior performance compared to baselines.