Investigating and Mitigating Degree-Related Biases in Graph Convolutional Networks