GLIMG: Global and local item graphs for top-N recommender systems

•Integrating the item graphs into an adapted semi-supervised learning model.•The combination of global graph and local graphs achieves better performance.•Such combination will not introduce the instability of local models.