CoRec: A co-training approach for recommender systems

In Recommender Systems, a large amount of labeled data must be available beforehand to obtain good predictions. However, labeled data are often limited and expensive to obtain, since labeling typically requires human expertise, time, and labor. This paper proposes a framework, named CoRec, which is based on a co-training approach that drives two recommenders to agree with each other's predictions to generate their own. We used three publicly available datasets from movies, jokes and books domains, as well as two well-known recommender algorithms, to demonstrate the efficiency of the approach under different configurations. The experiments show that better accuracy can be obtained when recommender algorithms are simultaneously co-trained from multiple views to make predictions.